Does Internal Control over Financial Reporting Curb Corporate Corruption? Evidence from China Weili Ge Michael G. Foster School of Business University of Washington Seattle, WA 98195 [email protected] Zining Li Cox School of Business Southern Methodist University Dallas, TX 75275 [email protected] Qiliang Liu School of Management Huazhong University of Science and Technology Wuhan, China, 430074 [email protected] Sarah McVay Michael G. Foster School of Business University of Washington Seattle, WA 98195 [email protected] July 8, 2015 Ge and McVay would like to thank the Moss Adams Professorship and Glen & Lucille Legoe Professorship, respectively, at the University of Washington for financial support. Liu acknowledges the financial support from the National Natural Science Foundation of China (Project No. 71332008 and 71172206). Ge and Liu acknowledge the financial support from the National Natural Science Foundation of China (Project No. 71328201 and 71332008). We thank Charles Lee, Bret Johnson, Joe Piotroski, Holly Skaife, Brandon Szerwo, Siew Hong Teoh, Tianyu Zhang, and workshop participants at Stanford University, the University of California at Davis, the City University of Hong Kong, the University of Missouri at Columbia, University of Toronto, Huazhong University of Science and Technology, Shanghai University of Finance and Economics, Southwestern University of Finance and Economics, Sun Yat-sen University, Yunnan University of Finance and Economics, and Zhejiang University for helpful comments. We thank the conference participants at the 2014 HKUST Accounting Research Symposium and the 2015 George Mason’s Investor Protection Research Conference for their helpful comments. We thank Professor Hanwen Chen and his team for sharing the internal control index data developed under National Natural Science Foundation of China Project No.71332008. Does Internal Control over Financial Reporting Curb Corporate Corruption? Evidence from China Abstract We examine whether internal control over financial reporting (internal controls) reduces the extent that managers or controlling shareholders expropriate resources from the firm. Using a sample of Chinese-listed firms, we find that after controlling for known determinants of corruption and internal control effectiveness, managers and controlling shareholders at firms with strong internal controls are significantly less likely to extract resources from the firm. We find this relation is concentrated within non-state-owned firms, where internal controls are less likely to be policy-driven, providing some evidence of the need to establish substance over form when assessing the strength of internal control. Taken together, our results suggest that, on average, strong internal controls curb corporate corruption; however, institutional factors and the associated incentives play a significant role in the effectiveness of internal controls. Keywords: Internal control over financial reporting; corporate corruption; state-owned enterprises I. INTRODUCTION We examine the relation between the strength of internal control over financial reporting (hereafter internal control) and corporate corruption. 1 We define corporate corruption as a form of corporate abuse in which controlling shareholders or managers expropriate resources from the firm and therefore minority shareholders.2 For example, CEOs are often persuaded through bribes to contract with inferior suppliers offering higher-priced or lower-quality inventory relative to other suppliers. CEOs can also transfer funds to related parties as loans with indefinite loan terms. Another common tactic for managers to extract resources from the firm is through the reimbursement for their private consumption expenses. Although corruption within Chinese firms has been well-documented (e.g., Jiang et al. 2010; Cai et al. 2011; Piotroski and Wong 2012), our interest is in whether internal controls can effectively combat this type of corruption. Similarly, although a number of studies we discuss in the next section examine the impact of internal controls on financial reporting and real activities, none have examined explicit evidence of corruption, or resource extraction; the 1 Internal control over financial reporting (ICFR) comprises the processes and procedures established by management to maintain records that accurately reflect the firm’s transactions, and covers asset representation, including asset misappropriation. Many of the same policies, procedures, and controls that lead to effective ICFR therefore also affect firm resources and operations. We focus on ICFR rather than all internal controls for several reasons. First, it is difficult to distinguish some broader internal controls from governance (e.g., an independent board). Second, ICFR is the focus of Section 404 of the Sarbanes-Oxley Act and the subsequent regulations within numerous countries, including China. Thus, our study provides evidence on whether these ICFR-focused regulations can be effective in curbing corruption. In later analyses we partition ICFR by operational controls and compliance controls, corroborating that it is operational ICFR that curb corruption. 2 We restrict our discussion of corruption to self-dealing by controlling shareholders or management extracting resources from the firm at the expense of minority shareholders. This is related to corruption within the government, given China’s unique institutional environment. Shleifer and Vishny (1993) define governmental corruption as “the sale by government officials of government property for personal gain.” Over half of Chinese publicly traded firms are state-owned firms (often termed as state-owned enterprises, SOEs), meaning that the government is the controlling shareholder. As a result, the government appoints individuals who are often current or former government bureaucrats as CEOs of state-owned enterprises. 1 closest provides indirect evidence of rent extraction via insider trading (e.g., Skaife et al. 2013). China is well-suited as a setting to examine the effect of internal control strength on corporate corruption. Unlike the U.S., within China, corruption is common and measurable. In particular, because the legal system offers only weak protection for minority investors, corporate abuse by controlling shareholders and managers is wide-spread and rather severe (Jiang et al. 2010) and there is a high risk of controlling shareholders expropriating resources from minority investors (Djankov et al. 2008). Common approaches of corruption documented in prior research are the tunneling of cash from the firm through loans, which are generally not repaid (Jiang et al. 2010), and the payment of private consumption expenses with firm resources (e.g., extravagant dinners or gambling expenses; Cai et al. 2011). We also consider ex post evidence of embezzlement or the receipt of bribes to enter into suboptimal contracts. Second, also distinct from the U.S., the required disclosures in China allow for a continuous measure of internal control strength across all firms,3 whereas only a small subset of firms with ineffective internal controls provide granularity on their underlying problems in the U.S. 4 This allows us to distinguish between the design of internal controls, and the implementation of internal controls. In particular, managerial override of existing internal controls is deemed as “ineffective internal control” in the U.S., whereas in China, the ICFR 3 We use a proprietary database that tracks listed Chinese firms’ internal control information from financial statements, filings to China Securities Regulatory Commission, government documents, and press releases (Chen et al. 2013). The database covers 99 percent of all Chinese public firms from 2007 through 2010 and allows us to measure the strength of internal control within the COSO framework. 4 For example, only 7.2% of firm-year observations disclose ineffective internal controls in the U.S. sample examined in Feng et al. (2015). The other 92.8% simply disclose they maintain effective internal controls, whereas our data exhibits variation in ICFR strength across the entire population. 2 score is the underlying existence of the internal controls, not whether these policies are followed, which allows us to separate form versus substance. We hypothesize that, to the extent that managers and controlling shareholders are unable to completely override internal control mechanisms in place, stronger internal controls will help mitigate the risk of corporate corruption as a stricter control and monitoring environment would make it more difficult for top management or controlling shareholders to engage in activities that are harmful to minority shareholders. For example, the creation and implementation of a policy for how loans are determined and which expenses should be reimbursed would reduce the likelihood of managers and controlling shareholders siphoning cash from corporate accounts. Similarly, if a policy required three bids for each inventory purchase, managers would be less able to receive bribes to contract with an inferior supplier. If, however, managers are able to ignore or override the internal control policies and procedures, the internal control system would not effectively deter corruption, and would merely be window dressing (form over substance). We study a sample of 4,775 firm-year observations from 2007 through 2010 of Chinese firms listed on the Shanghai or Shenzhen Stock Exchanges. We use the data from Chen et al. (2013) who collected 144 items related to various aspects of firms’ internal controls, which we discuss extensively in Section III and Appendices A and B. We first pare down the 144 items identified in Chen et al. (2013) to focus on internal controls over financial reporting (64 items). We next provide evidence on the validity of our measure of internal control strength by confirming that firms with higher internal control scores exhibit less earnings management, measured by the absolute value of discretionary accruals and the 3 incidence of financial statement restatements. This finding is consistent with that of Chen et al. (2013) using the broader measure and also with prior US-sample research examining ICFR. Following prior literature, we measure corporate corruption in the following ways: (1) other receivables scaled by total assets and related party loans scaled by total assets to capture the use of loans to tunnel funds out of the firm by controlling shareholders (Jiang et al. 2010), (2) travel and entertainment costs scaled by sales to capture expenses incurred for managers’ private consumption (e.g., eating, drinking, sports club membership, and travel; Cai et al. 2011), and (3) public ex post disclosures of corruption by top management.5 We find that, after controlling for known determinants of corruption and internal control strength, firms with strong internal controls appear to tunnel fewer resources out of their firms using loans and report lower travel and entertainment expenses, consistent with our prediction that internal controls mitigate corporate corruption. We do not, however, find a significant association between internal control strength and the frequency of disclosed corruption across our full sample. Next we examine whether the relation between internal control strength and corporate corruption varies with whether the internal controls were more likely to be voluntarily adopted or policy driven, proxied by state ownership structure (state-owned versus non-stateowned enterprises). 6 In 2006, in an effort to reform state-owned enterprises, the Chinese 5 Note that we remove the corruption cases in which managers bribe government officials to benefit the firm, for example to protect against government expropriation or facilitate government services (e.g., to receive necessary permits and licenses). We focus on corrupt behavior by management that is at the expense of minority shareholders. 6 An important ownership characteristic of China’s listed firms is state ownership, where the government is the controlling shareholder and appoints the top management team. State-owned enterprises (SOEs) comprise more 4 government issued numerous guidelines to underscore the importance of internal controls for listed firms and applied a great deal of pressure on state-owned firms to follow the guidelines in establishing specific internal control policies and procedures. Thus, although not mandated, this pressure might have led managers of SOEs to implement boilerplate internal control policies and procedures to satisfy government regulators, but without intent to enforce or follow these internal control policies and procedures.7 This institutional difference might allow us to assess substance versus form, to the extent that policy driven internal controls (those within state-owned enterprises) are less effective at curbing corporate corruption compared to voluntary adoptions (those within nonstate-owned firms), either because the controls are designed, but not implemented, or because managers simply override the controls. Consistent with at least some window dressing (i.e., form over substance), we find that, across all the corporate corruption measures, internal control strength is less effective at curbing corporate corruption within state-owned firms relative to non-state-owned firms.8 We conclude our analysis with several additional tests. First, we corroborate that our measures of corruption are self-dealing and represent resource extraction from the firm. than half of all firms listed on China’s stock exchanges (Piotroski et al. 2014). Managers of state-owned enterprises tend to be former government officials who face multiple—and potentially conflicting—objectives (e.g., political incentives versus incentives to maximize firm value). 7 Providing some evidence of this, we find that economic determinants of ICFR strength have lower explanatory power for ICFR strength among SOEs relative to non-SOEs (Appendix D). We also find that among SOEs, the establishment of strong internal control policies and procedures is positively associated with future CEO promotions (Section V). 8 This finding also allows us to mitigate concerns that it is the crackdown on corporate corruption by the Chinese government, rather than the internal controls per se, that have curbed corruption. If it were solely the governmental oversight curbing corruption, we would expect it to either be pervasive across SOEs and nonSOEs, or be concentrated among SOEs, where the government has the greatest influence. Instead we find that ICFR curbs corruption more among non-SOEs. We also include year fixed effects to further control for China’s recent focus on curbing corruption. Finally, in our additional analyses we examine cross-sectional variation in ownership structure within non-SOEs. 5 Consistent with this, we find evidence that corruption is negatively associated with future operating performance. Second, we document cross-sectional variation within non-SOEs. In particular, internal controls are more effective in curbing corrupt behavior when there is a greater balance of power among the top shareholders, measured by when there are more shares owned by the second to fifth largest shareholders. This finding is consistent with the notion that within some firms (in our study either SOEs or non-SOEs that do not have a balance of power among the top shareholders), internal controls may be easily over-ridden and thus not effective, even if they are established. In contrast, in firms with a more balanced ownership structure, internal controls are more effective. This could be because the board sets more rigorous internal controls, and/or managers and controlling shareholders are less able to override the controls. Third, we provide evidence that SOEs’ internal controls are influenced by governmental pressure in that a) they appear to be less based on the underlying firm economics, and b) managers implementing more internal controls in SOEs are more likely to be promoted. We also consider a two-stage least-squares (2SLS) estimation procedure to mitigate concerns of correlated omitted variables (such as management integrity) affecting both the choice to establish internal controls and the choice to extract resources. We find similar results when using the fitted value of ICFR strength. Our findings make several contributions to the literature. First, we document the link between internal controls and corporate corruption, which has not been documented to date. Second, our results highlight that internal controls are significantly less effective in curbing corporate corruption within state-owned enterprises. This suggests that institutional factors and the associated incentives play a significant role in the effectiveness of internal controls 6 (similar to the findings of Lin et al. (2012) regarding independent director regulation). The internal control disclosure requirements of the Sarbanes-Oxley Act in the U.S. have triggered similar reforms in many other countries. However, our findings suggest that improvements in internal controls might not have the intended benefits (e.g., curbing corporate abuse) across all firms in emerging markets, especially when both ownership structure is highly concentrated and protection for minority shareholders is lacking. II. BACKGROUND AND PREDICTIONS Internal control over financial reporting (herein internal controls) is comprised of the processes and procedures established by management to maintain records that accurately reflect the firm’s transactions (Deloitte and Touche 2004). 9 Researchers have examined various benefits of effective internal controls. Prior research has shown that higher internal control strength reduces the unintentional errors in financial reporting (Ashbaugh-Skaife et al. 2008; Doyle et al. 2007), leading to more accurate management forecasts as managers use more accurate financial inputs to form their forecasts (Feng et al. 2009), and improving investment decisions (Cheng et al. 2013) as well as firm operating efficiency and firm performance (Feng et al. 2015). These studies rely on the notion that the reports generated from a system with ineffective internal controls contain errors (often unintentional errors), and thus the information that management uses to create financial statements and make decisions is faulty. 9 Following the Committee of Sponsoring Organizations of the Treadway Commission (COSO) framework (2013), we define internal control as “a process, effected by an entity’s board of directors, management, and other personnel, designed to provide reasonable assurance regarding the achievement of objectives relating to operations, reporting, and compliance.” The focus of this paper is on internal control over financial reporting, but in our robustness checks, we also provide evidence on internal controls that are not related to financial reporting. 7 There is limited evidence, however, on whether internal controls reduce managers’ rent extracting behavior. Donelson et al. (2014) examine the association between internal control weaknesses and intentional financial distortion (fraud), but do not examine resource extraction. Skaife et al. (2013) provide some evidence of rent extraction in that the profitability of insider trading is greater in firms with ineffective controls. They suggest that a weak internal control environment provides managers with an information advantage, enabling them to profit from their private information by selling before stock price declines. They do not provide evidence, however, of the more direct resource extraction we consider. In particular, although selling personal shares at a profit is evidence of opportunism, it is both more indirect and less costly to shareholders than the more egregious and direct form of corruption we examine.10 This is especially salient given Skaife et al. (2013) find only limited evidence of managers managing earnings before selling their personal shares. The focus of our study is whether internal controls reduce the extent of corporate corruption, which is an escalated form of managerial rent extracting behavior that represents a much more direct and quantifiable cost to shareholders. The limited research on internal controls and resource extraction in the U.S. is not surprising. Strong and well-enforced investor protection in the U.S. constrains insiders’ ability to acquire private control benefits (Leuz et al. 2003; La Porta et al. 2000); as a result, U.S. firms exhibit significantly less corporate corruption, on average, than firms in countries 10 As an example, the CFO of South Airline Co., Liming Chen, tunneled cash from the company to related parties using numerous loans. For instance, he took a corporate loan of 30 million Chinese Yuan from China CITIC Bank, and then moved the money in the form of other receivables to a company controlled by his friend, Zhuangwen Yao. To return the favor, Zhuangwen Yao, gave Chen a BMW worth 700,000 Chinese Yuan and a house worth 2.25 million Chinese Yuan as gifts. In addition, Chen took bribes of over 53 million Chinese Yuan from various sources to enter into various contracts. 8 with poor investor protection. It follows that there is no strong pattern of, or evidence on, how insiders engage in corporate corruption in the U.S., making it difficult to measure corruption. Using a sample of Chinese firms allows us to overcome these limitations. China has particularly weak minority investor protection, in part because China’s legal system lacks enforcement.11 Allen et al. (2005) provide evidence that among seven developing countries, China’s corruption index is ranked as the most severe. 12 Thus, corrupt behavior by controlling shareholders and managers is wide-spread (Jiang et al. 2010). This corruption is also measurable. Prior research provides evidence on common approaches through which controlling shareholders or managers expropriate resources from minority shareholders in China (Jiang et al. 2010; Cai et al. 2011). Therefore, we are able to measure specific corporate corruption such as the tunneling of cash from the firm through loans, the payment of lavish entertainment expenses and other private consumption expenses with firm resources, or the ex post revelation that managers accepted bribes (for example to contract with inferior suppliers) or embezzled from the firm. We predict that it is more difficult for managers and controlling shareholders to extract resources from the firm in a stricter internal control environment. According to the government guidelines, the Board of Directors is in charge of establishing internal control systems. Therefore, to the extent that managers and controlling shareholders are unable to 11 According to the 2014 report by Word Bank Group, China was ranked 132 out of 189 countries in terms of the strength of minority shareholder protections. 12 These seven countries are: China, India, Pakistan, South Africa, Argentina, Brazil, and Mexico. Allen et al. (2005) base their inferences on the International Country Risk Guide’s assessment of the corruption in government. Lower scores suggest that “high government officials are likely to demand special payments” and “illegal payments are generally expected throughout lower levels of government” in the form of “bribes connected with import and export licenses, tax assessment, policy protection, etc.” 9 completely override internal control mechanisms in place, we expect internal controls to reduce the extent of their rent extraction behavior. For example, the common technique of issuing loans to transfer cash (i.e., tunneling) would be curbed by the requirement of a loan approval process that spells out interest and repayment terms. Related controls would trigger personnel to follow up on expected interest/principal payments that have not been received. With respect to private consumption by management, it is possible that requiring separate personnel to approve versus pay for invoices (i.e., segregation of duties) would curb much of the inappropriate reimbursement of private consumption expenses. In addition, beyond the establishment of segregation of duties, routine reviews of expense reports or maintaining a clear reimbursement policy would likely further reduce corrupt behavior of charging for items used for personal reasons or simply faking receipts. As another example related to accepting bribes, stricter purchase order authorization would ensure that managers are not over-ordering or ordering at an unreasonable price. We state our first hypothesis as the following, in the alternative form. H1: Strong internal controls reduce corporate corruption (resource extraction from the firm). Even in the presence of internal controls, managers or controlling shareholders may simply fail to implement or enforce these controls, and may also override the controls. Thus, if we do not find evidence consistent with H1, it could be because internal controls do not curb corruption, but it could also be due to the design of internal controls that are not effective (i.e., form over substance). Our next hypothesis should help provide evidence on the underlying reason for insignificant results. 10 Our second hypothesis examines whether the relation between internal control strength and corporate corruption varies with whether the internal controls were more likely to be voluntarily adopted or policy driven. To proxy for the construct of policy driven, we examine the ownership structure of the firm. Specifically, we expect that state owned firms in China receive more pressure to improve internal controls, and thus these controls are more likely to be policy driven than voluntary, which may influence the effectiveness of internal control in curbing corporate corruption. As of 2010, sixty-five percent of listed firms in China were state-owned enterprises, accounting for 89 percent of total market capitalization in China (Piotroski et al. 2015). The government is the controlling shareholder of state-owned enterprises and appoints key executives such as the CEO and the Chairman of the board. As a result, the top managers of state-owned enterprises have multiple objectives. In addition to profit maximization, they might also aspire to improve employment rates or build relationships with government superiors; these other objectives could cause significant inefficiencies for the firm (see Piotroski and Wong 2012; Piotroski et al. 2015). Following the implementation of the Sarbanes-Oxley Act in the U.S., the Chinese government began to emphasize improving publicly listed firms’ internal controls. In June 2005, the Ministry of Finance, the China Securities Regulatory Commission (CSRC), and the State-Owned Assets Supervision and Administration Commission (SOASAC) jointly issued the “Report on Learning from Sarbanes Oxley to Strengthen Our Listed Firms’ Internal Controls”. On March 5, 2006, Premier Jiabao Wen emphasized during the Fourth Plenary Meeting of the Tenth People’s Congress that “we need to introduce and learn from other 11 countries’ experiences in corporate governance, standardize governance mechanisms, and improve internal control systems.” A series of governmental guidelines were issued following Premier Wen’s address. The central government wanted to establish “role models” of stateowned enterprises to benefit the social goals of the government. The guidelines issued by SOASAC state that central-government-controlled enterprises should develop internal control systems and prevent corruption. Moreover, the “Basic Standards for Large and Median SOEs on Developing Modern Enterprises System and Strengthening Corporate Governance” issued by the General Office of the State Council of PRC states the following: “Those state-owned enterprises classified as major enterprises by the central and local governments, are required to identify deficiencies according to the Standards and make improvements to comply with the Standards. All other enterprises should also follow the Standards and strive to meet all the requirements. The committee on trade and economy, and the budget committee at the central government and local governments are called to conduct research and investigation on the effectiveness in implementing the standards.” Clearly, state-owned enterprises are under pressure from the government to establish certain types of internal control procedures, which is compounded by managers of the state-owned enterprises (current or former government bureaucrats) having incentives to maintain strong political connections with government officials. It is possible that state-owned firms may adopt a boilerplate list of internal control procedures to satisfy government regulators but do not actually implement or enforce these internal control policies and procedures. As a result, such policy-driven burdens (e.g., internal controls) may not be effective in achieving the stated goal of reducing corporate corruption. As suggested in Lin et al. (1998), policy burdens reduce the efficiency of state-owned firms’ operations. In addition, as argued in Piotroski and Wong (2012), greater state involvement in 12 an economy creates incentives for financial reporting opacity to hide the rent-seeking activities of politicians and related parties. In sum, although management of state-owned enterprises have incentives to adopt strong internal controls on paper, they likely have fewer incentives to actually realize the benefits of these internal control practices, relative to firms that voluntarily choose to establish strong controls. Because they have the power to override or simply ignore internal controls in place, the most corrupt managers are unlikely to be restrained by internal controls. As an example, Zhaolu Zhou, the CEO of Yunnan Copper Co., a state-controlled enterprise with a market capitalization of 14 billion Chinese Yuan, received at least 19 million Chinese Yuan in bribes to award certain contracts at the expense of shareholders, and did so by overriding internal controls.13 We thus hypothesize that policy-driven adoptions of internal control guidelines and procedures are less effective in reducing corporate corruption: H2: Strong internal controls are less effective at curbing corporate corruption when they are policy driven relative to when they are voluntarily adopted. We use the adoption of internal controls within state-owned enterprises to proxy for policydriven adoptions and conduct several tests to assess the validity of this proxy. III. DATA, SAMPLE, AND CORPORATE CORRUPTION MEASURES Internal Control Index and Sample Selection Our ICFR measure is based on the underlying data used in Chen et al. (2013). These data cover 99% of all Chinese listed firms from 2007 to 2010 and indicate whether 144 13 Specifically, Zhao acknowledged in his self-reflection report that he had too much power and was able to override the internal control systems within the firm. Self-reflection reports are a requirement of bureaucratic prisoners in China, who are required to detail their transgressions and the motivations for their actions. 13 specific firm features exist within each firm-year. These 144 firm features each fall within the five main aspects of internal control proposed by COSO: (1) Control Environment, (2) Risk Assessment, (3) Control Activities, (4) Information and Communication, and (5) Monitoring (see Appendix A).14 As our focus is ICFR, we focus on the existence of 64 of the 144 firm features collected for the initial index (see Appendix B). Each of the 64 items receives a value between zero and one, which we then average within each control aspect (three-digit level in Chen et al. 2013). 15 Finally we aggregate the values from each control aspect and calculate an average ICFR score ranging from zero to one. See Appendix B for a more detailed description of the calculation of our ICFR score. Appendix C provides the definitions for all of our variables. We present the sample selection procedure in Table 1, Panel A. Our sample begins with all Chinese firms listed on the Shanghai and Shenzhen stock exchanges with internal control index data from 2007 through 2010. We remove 120 firm-years in the financial industry as financial firms are under different internal control requirements issued by the People’s Bank of China and the China Securities Regulatory Commission. We then delete 14 These data were collected by the research team led by Hanwen Chen, supported by China NSF grant #71332008, and Ministry of Education Social Science Major Research grant #10JJD630003. Since its creation, the index has gained recognition from researchers, practitioners and regulators as the metric of Chinese firms’ internal control effectiveness. These data are considered the most comprehensive and authoritative detail on internal control by Chinese regulators and security market participants. For example, on June 11, 2010, all three of the most authoritative Chinese financial newspapers, the China Securities Journal, Shanghai Securities News, and Securities Times, featured articles introducing the index. The researchers who developed the internal control index annually publish the top 100 firms that have the highest internal control scores in the China Securities Journal. Deloitte highlighted this index on the website of its Corporate Governance Center; and the Public Company Monitoring Division of the Shanghai Stock Exchange also acknowledged that the index “has significant reference values for our efforts of monitoring internal control and corporate governance.” 15 Sixty-two of the 64 items receive a score of one if a certain feature is in existence and zero otherwise; the two remaining items receive a standardized score ranging from zero to one. Of the 64, 10 relate more to compliance than to operational controls; in our robustness checks we partition the score further into ICFR_OPERATIONS and ICFR_COMPLIANCE and find our results are concentrated within ICFR_OPERATIONS. We discuss this analysis in Section V. 14 1,401 firm-year observations that are traded on the small and medium-sized enterprises (SMEs) board because these firms are subject to different internal control and disclosure requirements (e.g., they are only required to provide internal control reports every other year). Of the remaining firms, 468 firm-years do not have necessary data for the analysis on the determinants of internal control, resulting in a sample of 4,775 firm-year observations. We obtain the information on firms’ stock prices, company financials, industry classification, ownership structure, auditors, and largest shareholders from the Chinese Stock Market and Accounting Research (CSMAR) database. As shown in Panel B of Table 1, our sample is evenly distributed across our sample period, and about 67% of our sample firms are stateowned enterprises, i.e., the controlling shareholders are either the central or local government, or their agencies. [Table 1] We validate our ICFR Score in two ways. We first explore the underlying determinants of our ICFR Score confirming that it exhibits expected associations with previously examined determinants of ICFR strength. We also confirm the association between ICFR and financial reporting quality documented in prior research (e.g., Doyle et al. 2007b; Chen et al. 2013) to further validate our internal control measure. Appendix D provides a detailed description of these validity tests. These associations help to mitigate concerns about measurement error in our proxy of ICFR strength. In particular, although some assessments require subjectivity, and others could reflect window dressing, both of these measurement errors should add noise and thus lower the power of our tests. That we find the expected associations with both determinants and known consequences of ICFR strength mitigates this measurement error concern. 15 Measures of Corporate Corruption The first form of corporate corruption that we analyze is tunneling through the use of loans with indefinite repayment terms (hereafter tunneling). Jiang et al. (2010) document that tunneling is a prevalent and persistent method used by controlling shareholders to expropriate funds from minority shareholders. 16 We use two measures to capture tunneling behavior. Following Jiang et al., we first use other receivables scaled by total assets (OTHREC). In addition, Chinese listed firms are required to disclose related party loans in their annual reports; thus we adopt a second measure (RELATEDLOAN) that is the amount of loans from the listed firm to disclosed related parties (including the controlling shareholder, other companies controlled by the same controlling shareholder, subsidiaries, and joint ventures) scaled by year-end total assets. We view OTHREC and RELATEDLOAN as complementary. OTHREC could capture the expropriation of funds via loans that are not disclosed as related party loans in annual report (e.g., funds to unidentified related parties) while RELATEDLOAN specifically captures lending behavior to related parties, but could miss other tunneling. At times, both measures will capture the same loan; the Spearman correlation between OTHREC and RELATEDLOAN is 0.36. Tunneling captures one specific approach through which controlling shareholders or managers can expropriate funds for personal gain. Prior research has also shown that corporate executives in China often misuse corporate funds for their private consumption such as dining, travel and entertainment. In China, entertainment providers generally are willing to provide generic or modified receipts for their clients, in order to facilitate 16 Clearly some of these loans are valid, with reasonable interest rates and standard repayment terms. These valid loans should serve to add noise, but not bias, to our analysis. We expect that the pervasiveness of tunneling has declined since the Jiang et al. study, but continue to find evidence of tunneling (see also footnote 21). We corroborate our inferences in Section V by documenting that this tunneling behavior is associated with poorer future performance. 16 reimbursement by their employers. Thus extremely lax accounting regulations and enforcement allow the reimbursement of private consumption expenses to be classified as business related administrative expenses. 17 We manually collect the information from the notes to financial statements in the annual reports on the following expenditures: office supplies, business travel, entertainment, communication, training abroad, board meetings, automobile, conferences, and other expenses. Thus, our second proxy of corruption is the sum of these expenditures scaled by sales (PRIVATE) which is intended to capture managers’ expropriation of corporate funds for private consumption (Cai et al. 2011).18 Our final measure of corruption is the ex post detection and disclosure of corruption by either regulators or the media. We collect exposed corruption cases from both main stream Chinese media sources and litigation cases to ensure the most comprehensive coverage.19 We require that this disclosure indicates self-serving behavior (e.g., embezzlement or the receipt of bribes) by management. We analyze each corruption case to determine the timing of corrupt behavior (i.e., when management undertakes the corrupt behavior), and set EXPOSED equal to one for the firm-years with the corrupt behavior, and zero otherwise.20 17 Anecdotal support of this behavior is common. As covered in Wall Street Journal on November 4th, 2014, the casinos in Macau (a Chinese territory) experienced their largest sales decline ever of 23% in October of 2014. “Industry executives and analysts attribute the recent poor performance in Macau to a variety of factors, particularly a Beijing-led crackdown on corruption that has caused VIP gamblers to shy away from the baccarat tables,” (O’Keeffe 2014). 18 PRIVATE could also capture the expenditures used to bribe external parties to negotiate deals that benefit the firm. It is likely that internal controls will not curb such bribing behavior as it benefits the firm; therefore, this might weaken the effect of internal control on PRIVATE. 19 To collect publicly disclosed corruption cases, we first search the “China Economic News Database,” which consists of articles published in all Chinese newspapers and periodicals. We use the following keywords (in Chinese) to conduct the news search: corruption, Shuanggui (detained and interrogated), stepping down, economic issues, embezzlement, misappropriation, bribery, job suspension, favoritism, property transfer, and Xiake (a synonym of stepping down). Next we searched within all legal case documents that are relevant to corruption from the courts in accordance with “LawInfoChina” (LawInfoChina is a data center for court legal documents). Finally, we used the Baidu search engine to retrieve public companies’ corruption materials based on company names identified from our search. 20 We do not include corruption cases that involve giving bribes to external parties (e.g., government) for the firm’s benefits (e.g., obtaining a contract with the government). Of the 159 cases involving exposed corruption, 120 cases relate to accepting bribes at the expense of shareholders (e.g., to buy inventory at above-market prices). The remaining cases involve embezzlement. Because some corruption cases involve multiple years, we have 214 firm-year observations with EXPOSED = 1. To minimize the effect of Type II errors on our analyses 17 We examine the association between internal control and corrupt behavior at the time the corruption occurred (not at the time of ex post disclosure). One advantage of this measure (EXPOSED) is that we have a high level of confidence regarding the existence of corrupt behavior (the type I error rate is low). These detected and exposed corruption cases tend to be egregious in nature. One disadvantage, however, is that many corruption cases likely remain unidentified (the type II error rate is high) and there could be selection biases in the exposed cases. In contrast, it is likely that our other measures (OTHREC, RELATEDLOAN and PRIVATE) have higher type I error rates, but lower type II error rates. We also note that tunneling behavior (OTHREC and RELATEDLOAN) is more likely to capture corrupt behavior by controlling shareholders while private consumption (PRIVATE) and accepting bribes (EXPOSED) is more likely to capture corrupt behavior by management. Thus, it is important to consider evidence from all of our corruption measures. To corroborate our main analysis, in Section V we provide evidence on the ex post performance consequence of corporate corruption; that is, whether greater corporate corruption is associated with lower future firm performance. We report descriptive statistics of the main variables in Panel A of Table 2. In our sample, the mean (median) of ICFR Score is 0.465 (0.460). This value is relatively sticky across time for the same firm, with an autocorrelation of 0.679. The mean (median) of OTHREC (tunneling) is 2.7% (1.2%) of total assets, where the mean (median) of total assets is 6,966 (2,504) million Chinese Yuan.21 The mean (median) of RELATEDLOAN is 1.2% (i.e., undetected corruption cases contaminating our control sample), we remove firms with other types of CSRC regulation violations or qualified audit opinions from the control sample. 21 The pervasiveness of tunneling was greater prior to the “crackdown” on these loans by the China Securities Regulatory Commission following the public disclosure of the evidence in Jiang et al. (2010). Thus, the mean of OTHREC in our sample is lower than that of Jiang et al. (2.7% versus 8.1%). Nonetheless, the extraction of resources via the tunneling of 2.7% of total assets remains material. To put this number in perspective, the mean of ROA is 3.6%. Further, we conducted a falsification test to examine whether ICFR influences accounts receivable the same way as other receivables. Unlike OTHREC, we do not find a negative association between 18 (0.0%) of total assets. The mean (median) of PRIVATE is 2.7% (1.0%), where the mean (median) of sales revenue is 4,935 (1,444) million Chinese Yuan. During our sample period, on average 4.5% of sample firms are detected and exposed for self-dealing corruption such as accepting bribes or embezzling assets. [Table 2] We report descriptive statistics by state-owned enterprises (SOEs) and non-SOEs in Panel B of Table 2. Consistent with the push by the government to establish internal control procedures within SOEs, the ICFR scores are higher for SOEs than non-SOEs at both the mean and median. For example, the mean score for SOEs is 0.475 versus 0.446 for non-SOEs. In Appendix D we explore the determinants of ICFR Score and see that SOEs continue to have higher ICFR after controlling for standard determinants of internal control strength, such as size and complexity. With respect to our corruption measures, tunneling (OTHREC and RELATEDLOAN) and reimbursement of private consumption expenses (PRIVATE) appear more pervasive among non-SOEs whereas there are notably more instances of publicly disclosed corruption within SOEs, with the mean value of EXPOSED equal to 0.062 for SOEs compared to a mean value of 0.011 for non-SOEs. Thus, there is some evidence that managers of non-SOEs are more likely to extract funds via loans that they do not expect to repay or through questionable expense reimbursements relative to managers of SOEs, who are more likely to accept bribes to enter suboptimal contracts, or embezzle from their firms. ICFR and accounts receivable. This result is consistent with the notion that OTHREC likely captures tunneling behavior that is beyond legitimate operating activities. We provide evidence in Section V on the potential negative consequence of tunneling during our sample period by examining how tunneling influences firms’ future performance to corroborate our inferences that OTHREC is costly to the firm. 19 SOEs are generally larger than non-SOEs, but are slightly less profitable, consistent with these firms and managers having objectives other than solely profit maximization. Finally, the Big Four audit firms have a much larger presence within SOEs, although the vast majority of firms are audited by smaller audit firms, consistent with prior research (e.g., Chen et al. 2011). Panel C of Table 3 presents the descriptive statistics of ICFR Score and our corruption measures by industry. It appears that the construction industry has the highest average ICFR Score (0.516) while the industry of agriculture, forestry and fishing has the lowest average ICFR Score (0.427). Generally it appears that although our corruption measures are populated across industries, our tunneling measures are most prevalent in Agriculture, Construction and Information Technology, PRIVATE is highest within Information Technology and Media (the latter perhaps reflecting a necessary cost in that industry) and EXPOSED is highest in Mining, Utilities, and Transportation. Panel D of Table 2 presents the means of ICFR Score and our corruption measures over time. In general, internal control strength has improved over time. Our corruption measures show a similar decline across time, consistent with improvements in ICFR curbing corruption. Turning to the correlation table in Panel E of Table 3, we find that ICFR Score is negatively correlated with OTHREC, RELATEDLOAN, and PRIVATE (consistent with H1), but not EXPOSED. From the low correlations of our corruption variables (except for the previously noted correlation between OTHREC and RELATEDLOAN), it is clear that each captures a distinct component of corruption. IV. INTERNAL CONTROL AND CORPORATE CORRUPTION Internal Control and Tunneling To explore whether internal controls curb tunneling, we estimate the following model: 20 OTHREC or RELATEDLOAN = β0 +β1ICFR +β2ASSET +β3ROA +β4TOP1SHR +β5INVCOMP +β6BOARD +β7CEOCHAIR + β8MARKETIZATION +β9BIG4 +β10EXCHANGE + ε (1) The dependent variable is either OTHREC or RELATEDLOAN, both of which are intended to identify the existence of tunneling (cash loans that are not expected to be repaid). OTHREC is other receivables scaled by year-end total assets and RELATEDLOAN is the amount of loans from the listed firm to related parties scaled by year-end total assets. ICFR is a variable ranging from zero to one with a higher value indicating stronger internal controls over financial reporting (see Appendix B). H1 predicts that effective internal controls reduce the extent of corporate corruption; therefore, we expect the coefficient on ICFR to be negative. Following Jiang et al. (2010), we include firm size (ASSET), profitability (ROA), shares owned by the largest blockholder (TOP1SHR), and the development of the regional market in which the firm is registered (MARKETIZATION) as control variables. Jiang et al. (2010) show that tunneling decreases in firm size and firm profitability. They also show that tunneling decreases in the ownership of the controlling shareholder because a large ownership interest mitigates the controlling shareholder’s incentives to extract private benefits that would destroy firm value. Tunneling is also expected to be more serious in a less developed regional market as less developed local markets are associated with weaker legal and regulatory environments. In addition to the controls included in Jiang et al. (2010), we include CEOCHAIR which equals one if the CEO is also the Chairman of the board, because we expect more powerful CEOs to have more opportunities to misappropriate corporate funds. We also include two control variables based on corporate governance items collected by Chen et al. (2013). Specifically, INVCOMP is a summary measure based on the four items related the shareholder composition (e.g., percentage of ownership by institutional investors) and BOARD is a summary measured based on the five items related to the characteristics of 21 the board of directors (e.g., the number of board committees). 22 These two variables are described in detail in Appendix C. We include BIG4 to control for differences in audit quality and EXCHANGE to control for different regulations and requirements under the two exchanges, although we do not have a directional prediction for this variable. Finally we include year and industry fixed effects in the regression. We report the regression results in Table 3 Panel A. Consistent with H1, the coefficient on ICFR is significantly negatively associated with both tunneling measures (e.g., coefficient = –0.013; p-value = 0.048 for OTHREC), suggesting that stronger internal controls curb tunneling. The coefficients on the control variables are consistent with those in Jiang et al. (2010). Specifically, tunneling is decreasing in firm size (ASSETS), profitability (ROA), percentage of shares owned by the largest shareholder (TOP1SHR), and the degree of development of regional markets (MARKETIZATION). We also find that higher institutional ownership and more active investors (INVCOMP) are associated with less tunneling behavior. We do not find a significant association between tunneling and the CEO also being the Chairman, nor between tunneling and audits by the Big Four. Next we examine H2 by estimating Equation (1) separately for SOEs and non-SOEs. The regression results are presented in Table 3, Panel B. Interestingly, for both tunneling measures, ICFR is no longer significant among SOEs (e.g., coefficient = –0.006; p-value = 0.207 for OTHREC), but is a strong predictor of tunneling within non-SOEs (e.g., coefficient = –0.032; p-value = 0.025 for OTHREC). The coefficient on ICFR is statistically different 22 This variable includes the number of dissenting proposals by independent directors, which offers a stronger measure of governance than independent directors in China, who may be ineffective (Lin et al. 2012). 22 between SOEs and non-SOEs (e.g., p-value = 0.068 for OTHREC).23 This result suggests that although internal control over financial reporting effectively mitigates tunneling within nonSOEs, this is not the case within SOEs. This result is consistent with our prediction that policy-driven adoptions of internal control are not as effective as adoptions that are more voluntary in nature (H2), highlighting the importance of substance over form. In summary, the results based on tunneling as a form of corporate corruption are consistent with both H1 and H2. We find a negative association between internal control strength and the extent of controlling shareholders harming minority shareholders by extracting resources from the firm via tunneling. Within SOEs, however, internal controls are not associated with this corrupt behavior.24 [Table 3] Internal Control and Private Consumption The preceding analysis provides evidence on the effect of internal control on corporate corruption specific to tunneling. Corporate corruption, however, manifests in various ways. As discussed in Section III, our next measure of corporate corruption is private consumption expense reimbursement (PRIVATE). Thus, we estimate the following model, where the dependent variable is PRIVATE. 23 To test whether the main coefficients are the same across different SOE types, we use the following Z- statistics: ܼ ൌ ିೕ ට௦మ ሺ ሻା௦మ ሺೕ ሻ ; where bi and bj are coefficient estimates from the two sub-samples, and s2(b) is the squared standard errors of the coefficients (Clogg et al. 1995, Chen et al. 2010). 24 This could reflect that ICFR does not constrain tunneling, but could also reflect that tunneling is not present in SOEs. To explore which of these is more descriptive, we consider the “costs” to tunneling on future operating income and find that tunneling is associated with lower future operating income among SOEs, providing some evidence that tunneling is present among SOEs (see Section V). 23 PRIVATE= β0 +β1ICFR +β2PAYRATIO +β3MGMTSHR +β4ATO +β5ROA + β6LOSS +β7EISSUE +β8DISSUE +β9BIG4 +β10TOP1SHR +β11INVCOMP +β12BOARD +β13CEOCHAIR +β14MARKETIZATION (2) PRIVATE is the sum of expenditures for office supplies, business travel, entertainment, communication, training abroad, board meetings, automobile, conferences, and other expenses scaled by sales. We predict a negative coefficient on ICFR because stronger internal controls should limit the magnitude of private benefits resulting from the corruption. We include several control variables from prior research. First, executives with lower pay are more likely to misappropriate corporate funds and assets for private consumption. Chen et al. (2005) find that executives substitute lower pay with excessive perquisite consumptions. We follow their measure of executive compensation and control for the ratio of executive total compensation relative to other employees’ average compensation (PAYRATIO). We also control for management ownership (MGMTSHR), measured as outstanding shares owned by executives, because management ownership helps align managers’ objectives with shareholders’ interests and reduces incentives for excessive perquisite consumption. We next include asset turnover (ATO) as a control variable to control for the effect of operating efficiency on selling, general and administrative expenses. We control for firm performance by return on assets (ROA) and whether there is a loss (LOSS) but we do not have a clear prediction on these two variables. On one hand, more profits provide more resources for corrupt executives; on the other hand, uncorrupt managers might generate better performance (e.g., they will buy the best-value inventory). We control for new issuances of equity (EISSUE) and debt (DISSUE) since the expenditures included in our PRIVATE measure may 24 increase during financing activities. Lastly we control for BIG4, TOP1SHR, INVCOMP, BOARD, CEOCHAIR, and MARKETIZATION as in the tunneling analysis. Table 4 presents the regression results for PRIVATE, our proxy for private consumption expense reimbursement. We find a significant negative association between ICFR and PRIVATE for the full sample (coefficient = –0.025; p-value = 0.047), suggesting ICFR effectively reduces corruption related to private consumption, again supporting H1. We again find that ICFR is insignificant within SOEs (coefficient = –0.002; p-value = 0.459), but is a strong predictor of corruption within non-SOEs (coefficient = –0.069; p-value = 0.027). The coefficient on ICFR within non-SOEs is statistically and economically larger in magnitude than the coefficient on ICFR within SOEs, in support of H2.25 [Table 4] Internal Control and Exposure of Corrupt Behavior Our final measure of corporate corruption is disclosed corruption cases (EXPOSED). We estimate the following model, where the dependent variable is EXPOSED: EXPOSED= β0 +β1ICFR +β2PAYRATIO +β3MGMTSHR +β4SALES +β5ROA +β6EISSUE +β7DISSUE +β8TOP1SHR +β9INVCOMP +β10BOARD +β11CEOCHAIR +β12MARKETIZATION + ε (3) EXPOSED equal to one if we identify an exposed case of corrupt self-dealing (e.g., embezzlement or the receipt of bribes to conduct suboptimal transactions) from either main stream Chinese media sources or litigation cases. In addition to the corruption determinants included previously, we also include SALES as a control variable because a majority of the exposed cases (120 out of 159 cases) relate to accepting bribes (e.g., to buy inventory at 25 We also examine the robustness of this result by adopting a two-stage approach. First, we model PRIVATE by regressing PRIVATE on the potentially legitimate determinants (i.e., SOE, firm age, growth, leverage, size, ROA, the number of employees, and year and industry fixed effects) and then we estimate Equation (2) by using the residuals from the first stage regression. The results are inferentially similar (not tabulated). 25 above-market prices). CEOs of firms with more sales might have more opportunities to conduct this type of corruption (e.g., more suppliers will vie for the opportunity to bribe these managers as the gain is larger). We predict a negative coefficient on ICFR if stronger internal controls limit the opportunity for corporate corruption. We provide the regression results for Equation (3) in Table 5. The association between ICFR and EXPOSED is insignificant for the full sample and the SOE sample, and is marginally significantly negative for the non-SOE sample. This is generally consistent with our results on tunneling and private consumption in that internal controls appear to be effective at reducing corruption within firms owned and controlled by entrepreneurs and private investors, but not within state-owned enterprises. 26 We find a significant negative association for the level of executive pay relative to other employees (PAYRATIO) within the full, SOE and non-SOE samples, suggesting that a lower relative executive pay creates incentives for managers to accept bribes or embezzlement. Management ownership seems to reduce the incentives to engage in corruptions as results indicate a positive association between EXPOSED and MGMTSHR for the full and non-SOE samples. We also find some evidence that powerful CEOs (CEOCHAIR = 1) are more likely to accept bribes or embezzle within non-SOEs. [Table 5] Taken together, our results from Tables 3–5 provide evidence that the strength of internal controls over financial reporting curbs corporate corruption (H1). Internal controls 26 Exposed bribery cases are concentrated in SOEs, which is consistent with the larger size and greater scrutiny of SOEs. This bias should serve to work against us finding an association between ICFR and EXPOSED among non-SOEs. 26 within state-owned enterprises, however, are much less effective at curbing corruption across all four of our tests, consistent with the importance of substance over form (H2). V. ADDITIONAL ANALYSES AND ROBUSTNESS TESTS Corroboration of Corruption Measures An underlying assumption in our prior tests is that our measures of corruption capture expropriation of resources from the firm and minority shareholders. If such corrupt behavior truly harms the firm, then we expect to observe worse ex post firm performance for firms exhibiting more corrupt behavior. Thus, we expect our measures of corruption to be associated with lower future firm performance. We estimate the following model, where the dependent variable is one-year-ahead operating income scaled by average total assets (OIt+1): OIt+1= β0 +β1Corruption measures +β2PPE +β3 GROWTH +β4SEGMENTS +β5EXPORT (4) +β6AGE +β7BIG4 +β8OI +β9ASSETS +β10TOP1SHR + ε We base our control variables on those considered by Feng et al. (2015), who examine ex post performance following ineffective internal controls (see their Table 8), and include TOP1SHR following Fan et al. (2007). The results are reported in Table 6. We find that OTHREC, RELATEDLOAN, and EXPOSED all have a significantly negative association with one-year-ahead operating income. These results suggest that our corruption measures do capture management behavior that is detrimental to subsequent firm performance. PRIVATE, however, is not significant in predicting one-year-ahead operating income. Among the control variables, we find that future operating income is higher for firms with high past operating income and a Big Four auditor, and, similar to the results in Feng et al., (2015), 27 future operating income is decreasing in capital intensity (PPE) and sales growth (GROWTH).27 [Table 6] The Balance of Shareholder Power and the Effectiveness of Internal Controls In this section, we examine how the balance of shareholder power influences the effectiveness of internal controls. Prior research has suggested that the primary agency problem in China is the risk that controlling shareholders expropriate resources at the cost of minority shareholders (Jiang et al. 2010). Thus, we explore whether internal controls are more effective in curbing corrupt behavior when there is a greater balance of power among major shareholders. In particular, we partition firms by the proportion of shares held by the second to the fifth shareholders. A higher value would suggest that there is a greater balance of power among major shareholders. In these firms, we expect internal controls to be more effective, because the board initially sets more rigorous controls, and/or because controlling shareholders and management are less able or willing to override these controls. In sum, we expect internal controls to be more effective in curbing corrupt behavior when there is a greater balance of power among shareholders. For our full sample, the mean (median) shares owned by the second to fifth largest shareholders is 10.4% (7.5%) (Table 2 Panel A). This percentage is significantly lower for stated-owned firms (mean = 10.0% and median = 6.6%), compared with that of non-statedowned firms (mean = 11.2% and median = 9.3%) (Table 2 Panel B). To examine whether the effectiveness of ICFR is higher when there is a greater balance of power among major 27 The results are similar when we estimate Equation (4) separately for SOEs and non-SOEs; that is, for both SOEs and non-SOEs, OTHREC, RELATEDLOAN, and EXPOSED all have a significantly negative association with one-year-ahead operating income. 28 shareholders, we first create an indicator variable TOP2to5IND that takes the value of one if the ownership by the second to fifth largest shareholders is greater than or equal to the full sample median (i.e. 7.5%), and zero otherwise. We then add TOP2to5IND and the interaction term, ICFR×TOP2to5IND, to Equations (1)-(3) and repeat our main analyses. We expect a negative coefficient on ICFR×TOP2to5IND if internal controls are more effective in curbing corrupt behavior when TOP2to5 is above the median. The results are reported in Table 7. As shown in Panel A of Table 7, for the full sample we find that ICFR×TOP2to5IND is significantly negative in the regressions of OTHREC and RELATEDLOAN, but not in the regressions of PRIVATE and EXPOSED. Turning to Panel B of Table 7, when we consider SOEs and non-SOEs separately, we find ICFR×TOP2to5IND is significantly negative across all measures of corruption within the non-SOE sample. Each of the coefficients on ICFR×TOP2to5IND, however, is insignificant within the SOE sample across all four measures of corruption. The differences between the coefficients of ICFR×TOP2to5IND for the SOE and non-SOE samples are also statistically significant across all four corruption measures. Overall, these results are consistent with our prediction that internal controls are more effective when there is a greater balance of power among the major shareholders. That we only observe this relation within non-state-owned firms is consistent with our H2.28 [Table 7] 28 An alternative partitioning variable is family firms, which comprise 80 percent of the non-SOEs in our sample. We do not focus on this partition for two reasons. First, given the high frequency of family firms, this partition would likely lack sufficient variation to be meaningful. Second, the variation we do examine (shareholdings by the second to fifth shareholder) also serves as a partitioning variable within family firms. 29 Corroboration of Governmental Pressure on SOEs Our earlier results are consistent with H2 that internal controls are not as effective in curbing corporate corruption within SOEs compared to non-SOEs. We motivate this hypothesis by positing that SOE management have different incentives to adopt internal control practices than non-SOE management, resulting in internal controls in SOEs that are effectively “window-dressing.” We provide preliminary support of this in Appendix D where SOEs tend to have stronger ICFR than non-SOEs, after controlling for known determinants of ICFR. To provide further evidence on this, we examine whether internal control strength is associated with the likelihood of CEOs receiving promotions. We identify 788 CEO turnovers and estimate the following model, where the dependent variable is PROMOTION. PROMOTION= β0 +β1ICFR +β2SALE +β3 OCF +β4ROA +β5LEV +β6MB +β7CEOAGE +β8EDUCATION +β9TENURE +β10TOP1SHR +β11MARKETIZATION + ε (5) We consider the CEO as having been promoted if in the year subsequent to her departure she became (1) the Chairman of the board at the same company, (2) the CEO or the Chairman/Vice Chairman of the board in the parent company, (3) the CEO at a larger public company where the assets of the new company are at least 20% larger than the former company, or (4) a high-ranking government official.29 We first estimate Equation (5) using an Ordered Logit regression including all firmyears. The dependent variable takes the value of one if the CEO is promoted in the following year, zero if there is no CEO turnover, and –1 otherwise. Next we estimate a Logit regression including only firm-years with CEO turnovers. The dependent variable takes the value of one 29 Executive positions at Chinese SOEs tend to have an “equivalent rank” as that of positions within the government, which allows us to assess whether a new governmental position is a promotion. For example, Yongkang Zhou, CEO of Chinese Petroleum had an equivalent rank of the Vice Minister of the State Owned Assets and Resources from 1996 to 1998. In 1998, he was promoted to the Minister of the State Owned Assets and Resources. 30 if the CEO is promoted in the following year, and zero otherwise. The results and the definitions of control variables are presented in Table 7. For the full sample, the coefficient on ICFR is significantly positive (p-value=0.016). The coefficient on ICFR is significantly positive (p-value=0.026) for SOEs; in contrast, ICFR is not associated with PROMOTION within non-SOEs.30 We find similar results using a Logit estimation within firms with CEO turnover. This finding suggests implementing strong internal controls improves career outcomes for SOE management. This is consistent with SOE management experiencing governmental pressure to establish certain internal control procedures, and corroborates our earlier results suggesting that the multiple objectives of SOE CEOs likely contribute to less effective internal controls (form over substance). [Table 8] Endogeneity of Internal Controls In this section we discuss the potential endogeneity issue that arises because internal controls are not exogenous; instead managers and the board of directors choose to establish internal control policies and procedures. Managers with a tendency to misappropriate assets may be more likely to choose weaker internal controls. To shed some light on the direction of causality, we regress all four lagged corruption variables on current ICFR and find no relation, mitigating the concern of reverse causality (p-values ranging from 0.189 to 0.829; not tabulated). We then repeat our main analyses on corruption using last-year’s ICFR instead of current year’s ICFR. We continue to find a significant negative association between lagged ICFR and corruption using OTHREC, RELATEDLOAN, and PRIVATE, and the effects 30 We note that PROMOTION for SOEs does not have the exact same meaning as PROMOTION for non-SOEs because only SOEs’ CEOs have the opportunity to be promoted to positions as government officials. 31 continue to be stronger among non-SOEs than SOEs. Collectively, these additional results support our inferences that the strength of internal controls over financial reporting curbs corporate corruption, but internal controls within state-owned enterprises are less effective.31 In addition, the choice of internal controls is affected by firm characteristics such as firm complexity and available resources. It is also possible that both corruption and internal controls stem from firm characteristics, but that internal controls, per se, would not curb corruption. One obvious correlated omitted variable might be China’s crackdown on corruption over this time period. Were it the government, rather than ICFR curbing corruption, we would expect it to either be pervasive across SOEs and non-SOEs, or be concentrated among SOEs, where the government has the greatest influence. Instead we find that ICFR curbs corruption more among non-SOEs. Further, our inclusion of year fixed effects should also help to capture China’s recent focus on curbing corruption. To investigate whether endogeneity is a concern more generally, we conduct a Hausman test following Larcker and Rusticus (2010) to test whether endogeneity is likely to exist. Specifically, we include both ICFR and the residual from the determinants model as described in Appendix D (the first stage model) in the second stage estimation of corruption (Equations 1–3). We find that the residual is significant in explaining each of our corporate corruption variables except for EXPOSED. Thus the test rejects the null of no endogeneity. 31 To further alleviate the concern of unidentified correlated omitted variables, we conduct a survey questioning management as to the link between internal controls and resource extraction. We randomly select 50 firms from our sample (24 SOEs and 26 non-SOEs) and administer a survey through the Accounting Office of the Shenzhen Stock Exchange. Our survey response rate is 100%. When asked whether effective internal controls can curb corrupt management behavior, 72 percent of the respondents strongly agreed. We do not find a statistically significant difference in the responses between SOEs and non-SOEs. This response is consistent with the notion that strong internal controls curb corporate corruption and reduce resource extraction. We view that this as complementing our empirical analyses and alleviating the concern about correlated omitted variables. 32 We next follow prior research and employ a two-stage least-squares (2SLS) estimation procedure (e.g., Gul et al. 2009). Our instrumental variable (or exclusion restriction variable per Lennox, Francis, and Wang (2012)) in the first stage regression is LIST2006, which we expect to be significantly associated with ICFR but for which we do not have ex ante reasons to believe would directly affect corporate corruption (except indirectly through ICFR). Statistically, in the first stage regression, we find that internal control strength is significantly associated with LIST2006. When we include LIST2006 in the second stage regressions, it is insignificant in explaining our three corruption variables. Therefore, our instrumental variable appears to be significantly correlated with ICFR, but related to our corporate corruption variables mainly through ICFR. As pointed out in Lennox et al. (2012), results based on the two-stage approach tend to be sensitive to the choice of the instrumental variable; in many cases the OLS approach is preferable. Therefore, we consider OLS in our main analysis but also confirm that our results are robust to using the two-stage approach. We repeat all of our analyses using the predicted internal control scores from the first stage regression and report the summary of results in Table 8. Our results are substantively similar for all four proxies of corporate corruptions (OTHREC, RELATEDLOAN, PRIVATE and EXPOSED); that is, firms with higher internal control scores have fewer other receivables, loans issued to related parties, and fewer expenses susceptible to private consumption for the full sample, and the association is significantly weaker within SOEs When examining EXPOSED as the dependent variable, we again only find evidence that internal controls are negatively associated with ex post disclosures of corrupt behavior within 33 non-SOEs. Taken together, these findings mitigate the concern that a particular firm feature drives both corruption and internal control strength, but that internal control strength alone would not curb corruption. [Table 9] Robustness Checks We conduct two robustness checks. We first consider whether the inclusion of internal controls related to other aspects but not financial reporting affects our results. To do so we select items related to internal control procedures that are not related to financial reporting (e.g., whether the firm has a legal counselor) and use the weighted average as the measure of internal control unrelated to financial reporting (ICDIFF). We then include ICDIFF as an additional control in our main regressions of corruption (Equations 1 – 3). Similar to our main results, we find the coefficient on ICFR remains significantly negative for the regressions of OTHREC, RELATEDLOAN, and PRIVATE (both full and non-SOE samples) and EXPOSED (non-SOE sample only), while the coefficient on ICDIFF is insignificant in these regressions for the full sample and both sub-samples. The lack of result for ICDIFF could be driven by the fact that firms exhibiting strong internal controls over financial reporting tend to have strong other internal controls, which is evident by a high correlation of 43% between ICDIFF and ICFR. Overall our results are similar when we control for internal controls other than financial reporting. Next, we investigate whether the effect of internal control strength on corporate corruption varies by the type of internal control procedures. We split the components in our ICFR measure into controls related to operations (ICFR_OPERATIONS) (e.g., controls for authorizing and approving) and controls related to compliance with regulations 34 (ICFR_COMPLIANCE) (e.g., having a system governing information disclosures). We identify 10 out of the 64 components that are more related to compliance than operations. We note that in many cases it is difficult to separate compliance controls from operational controls, and they are likely highly correlated. In fact, the Pearson correlation between these two types of controls is 0.508. We replace ICFR with ICFR_OPERATIONS and ICFR_COMPLIANCE in our main regressions (Equations 1 – 3). Similar to our main results, we find that, across all four corruption measures, firms with strong operational internal controls exhibit less corruption, and this association is concentrated within non-state-owned firms. The coefficient estimates on ICFR_COMPLIANCE, however, are insignificant across all specifications. This result suggests that the effect of internal control strength on corporate corruption is primarily driven by operational controls. VI. CONCLUSION We examine the relation between the strength of internal control over financial reporting and corporate corruption. Using a sample of Chinese firms, we find that after controlling for known determinants of corruption and internal control strength, firms with strong internal controls exhibit significantly less corporate corruption. Specifically, high internal control scores are associated with significantly less tunneling of resources out of the firms and lower travel and entertainment costs (our proxy for private consumption expense reimbursement). We also use the setting of state ownership to explore window dressing (form over substance) in the internal control arena. Because SOEs experienced intense governmental pressure to establish internal controls, we expect that these internal controls will be more 35 likely to be form over substance, as the management was pressured to adopt the policies. Thus, we explore the differences in the effectiveness of controls in mitigating corruption between SOEs and non-SOEs. We find that the state ownership of Chinese firms significantly weakens the effect of internal control strength on the extent of corporate corruption in the forms of tunneling resources and private consumption. In addition, we find that internal control strength has a significant negative association with the frequency of disclosed corruption within non-state-owned firms, but not state-owned firms. This finding is consistent with state-owned enterprises adopting internal control policies and procedures in form, but not substance. In summary, our results suggest that, on average, strong internal controls seem to reduce the extent of corporate corruption; however, institutional factors and the associated incentives also play a significant role in the effectiveness of internal controls. Internal control systems that are only form, without substance, are less effective. 36 References Allen, F., J. Qian, and M. Qian. 2005. Law, finance, and economic growth in China, Journal of Financial Economics 77, 57–116. Ashbaugh-Skaife, H., D. Collins, and W. Kinney. 2007. The discovery and reporting of internal control deficiencies prior to SOX-mandated audits, Journal of Accounting and Economics 44, 166–192. Ashbaugh-Skaife, H., D. Collins, W. Kinney, and R. LaFond. 2008. The effect of internal control deficiencies and their remediation on accrual quality. The Accounting Review 83(1), 217–250. Cai, H., H. Fang, and L. Xu. 2011. Eat, drink, firm and government: An investigation of corruption from the entertainment and travel costs of Chinese firms. Journal of Law and Economics 54, 55–78. Chen, H., W. Dong, H. Han, and N. Zhou. 2013. A comprehensive and quantitative internal control index: construction, validation, and impact. Working paper, Xiamen University, Zhejiang University, and State University of New York at Binghamton. Chen, D., X. Chen, and H. Wan. 2005. Compensation regulation in SOEs and excessive consumption of perquisites. Economics Research (Chinese) 2005(2). Chen, S., S. Sun, and D. Wu. 2010. Client importance, institutional improvements, and audit quality in China. The Accounting Review 85 (1), 127–158. Cheng, M., D. Dhaliwal, and Y. Zhang. 2013. Does investment efficiency improve after the disclosure of material weaknesses in internal control over financial reporting? Journal of Accounting and Economics 56, 1–18. Clogg, C., E. Petkova, and A. Haritou. 1995. Statistical methods for comparing regression coefficients between models. American Journal of Sociology 100 (5), 1261–1293. Dechow, P., R. Sloan, and A. Hutton. 1996. Causes and consequences of earnings manipulation: An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13, 1–36. Djankov, S., R. La Porta, F. Lopez-de-Silanes, and A. Shleifer. 2008. The law and economics of self-dealing. Journal of Financial Economics 88, 430–465. Donelson, D., M. Ege, and J. McInnis. 2014. Internal control weakness and financial reporting fraud. Working paper. University of Texas and University of Florida. 37 Doyle, J., W. Ge, and S. McVay. 2007a. Accruals quality and internal control over financial reporting. The Accounting Review 82, 1141–1170. Doyle, J., W. Ge, and S. McVay. 2007b. Determinants of weaknesses in internal control over financial reporting. Journal of Accounting and Economics 44, 193–223. Fan, G., and X. Wang. 2006. In: The report on the relative process of marketization of regions in China. The Economic Science Press, Beijing (in Chinese). Fan, J., T. J. Wong, and T. Zhang. 2007. Politically connected CEOs, corporate governance, and post-IPO performance of China’s newly partially privatized firms. Journal of Financial Economics 84, 330-357. Feng, M., C. Li, and S. McVay. 2009. Internal control and management guidance. Journal of Accounting and Economics 48, 190–209. Feng, M. C. Li, S. McVay and H. Skaife. 2015. Does ineffective internal control over financial reporting affect a firm’s operations? Evidence from firms’ inventory management. The Accounting Review 90(2), 529–557. Gul, F., S. Fung, and B. Jaggi. 2009. Earnings quality: some evidence on the role of auditor tenure and auditors’ industry expertise. Journal of Accounting and Economics 47, 265-287. Jiang, G., C. Lee, and H. Yue. 2010. Tunneling through intercorporate loans: the China experience, Journal of Financial Economics 98, 1–20. Kothari, S., A. Leone, and C. Wasley. 2005. Performance matched discretionary accrual measures. Journal of Accounting and Economics 39, 163–197. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and R. Vishny. 2000. Investor protection and corporate governance. Journal of Financial Economics 58, 3–27. Larcker, D. and T. Rusticus. 2010. On the use of instrumental variables in accounting research. Journal of Accounting and Economics 49(3), 186-205. Lennox, C., Francis J. and Z. Wang. 2012. Selection models in accounting research. The Accounting Review 87 (2), 589-616. Leuz, C., Nanda, D., and P. Wysocki. 2003. Earnings management and investor protection: An international comparison. Journal of Financial Economics 69, 505-527. Lin, J., F. Cai and Z. Li. 1998. China’s economic reforms: some unfinished business. American Economic Review 88(2), 422–427. 38 Lin, K., J. Piotroski, Y.G. Yang, and J. Tan. 2012. Voice or exit: Independent director decisions in an emerging economy. Working paper. Liu Q., L. Luo, W. He, and H. Chen. 2012. State ownership, institutional environment and internal control. Accounting Research (Chinese) 3, 52-61. Marinovic, I. 2013. Internal control system, earnings quality, and the dynamics of financial reporting. RAND Journal of Economics 44(1), 145–167. Okeeffe, K. 2014. Macau Gambling Revenue Falls 23% in October. Wall Street Journal. http://online.wsj.com/articles/macau-gambling-revenue-falls-23-in-october-1415079292 Piotroski, J., and T. J. Wong. 2012. Institutions and information environment of Chinese listed firms. Chapter from Capitalizing China (NBER), 201-242. Piotroski, J., T.J. Wong, and T. Zhang. 2015. Political incentives to suppress negative financial information: evidence from Chinese listed firms. Journal of Accounting Research 53(2), 405-459. Skaife, H., D. Veenman, and D. Wangerin. 2013. Internal control over financial reporting and managerial rent extraction: Evidence from the profitability of insider trading. Journal of Accounting and Economics 55, 91-110. Shleifer, A. and R.W. Vishny. 1993. Corruption. The Quarterly Journal of Economics, 599– 617. 39 Appendix A: Overall Internal Control Index This appendix provides a brief summary of the overall internal control index. For details about the index, please see Appendix 1 in Chen et al. (2013). A research team led by Professor Henwen Chen at the Xiamen University in China created the internal control index. The construction of the internal control index is based on the Basic Standards and the Guidelines for Chinese public companies,32 and the requirements issued by the Shenzhen and Shanghai Exchanges. 33 The design of the index follows the Internal Control—Integrated Framework (released by the Committee of Sponsoring Organizations of the Treadway Commission in 1992). Under this framework, the internal control index contains five components: control environment, risk assessment, control activities, information and communication, and monitoring. The COSO framework has gained acceptance by regulators, such as the SEC and PCAOB, and forms the foundation of SOX 404. The researchers adopted the analytic hierarchy process (AHP) developed by Thomas L. Saaty in the 1970s.34 Internal control index begins with five first-level items, the same as the five components in the COSO framework. Each of the first-level items then contains more evaluation items. For example, the codes for the first item in the first three levels are IC1, IC11 and IC111, respectively. For the fourth level, the code is defined as, for example, IC11101. The first three numbers (111) represent the first three levels to which this item is affiliated. If there is no third level for a particular item, then the first two numbers are presented. Overall, there are five first-level items, 24 second-level items, 43 third-level items, and 144 fourth-level items. In evaluating each of the 144 items regarding internal controls of a firm, the researchers collect its internal control information from financial statements, CSRC filings, government documents, and press releases. In most cases, a value of zero or one is assigned to a fourth-level item based on the information obtained (see Appendix B for examples). The researchers follow AHP to assign weights for the items (see Chen et al. for details on calculating the item weights) and the overall index score is the weighted average of all items. We identify 64 of the144 items that most directly relate to ICFR. See Appendix B. 32 On July 15, 2006, five Chinese government authorities and regulatory bodies, the Ministry of Finance, the China Securities Regulatory Commission, the National Audit Office, the China Banking Regulatory Commission, and the China Insurance Regulatory Commission jointly established a Committee aiming to stipulate a set of universal, recognizable, and scientific rules governing firms’ internal controls. After two years of conducting research and seeking feedback, the Committee on Internal Control Standards issued The Basic Standards of Enterprise Internal Controls (the Basic Standards) requiring that a listed firm issue a selfassessment of its internal controls and that a Certified Public Accountant issue a report on the firms’ internal controls. Later, Supplemental Guidelines of Firms’ Internal Controls (the Guidelines) was released on April 26, 2010. The Basic Standards became effective on January 1, 2011 for firms listed both on an exchange in China and another exchange abroad (dual-listed), and on January 1, 2012 for firms listed on the Shanghai Stock Exchange or the main section of the Shenzhen Stock Exchange. 33 In July 2006, the Shanghai Stock Exchange (SSE) issued A Guideline of Internal Controls for Listed Firms on Shanghai Stock Exchange. Further, in September 2006, the Shenzhen Stock Exchange issued A Guideline of Internal Controls for Listed Firms on Shenzhen Stock Exchange. 34 AHP decomposes a complex decision problem into a hierarchy of more easily comprehensible sub-problems, and then produces the qualitative and quantitative analyses of each sub-problem. AHP also makes pairwise comparisons of the same-level items in every sub-hierarchy, analyzing their relative impact on an element in the hierarchy above them. 40 Appendix B: Composition of the Internal Control over Financial Reporting Measure Out of the 144 items of the internal control index, we select only those items that apply to internal control over financial reporting. We remove 68 items that are not directly related to internal control but instead related to corporate governance (e.g., percentage of institutional shareholders), or items reflecting the outcome of internal controls (e.g., whether the company reported inventory damages in the year). We classify another 12 items as relating to overall internal controls but not financial reporting. As a result, our ICFR score consists of 64 evaluation items. The following provide a complete list of these components. Standardized values are calculated as the actual value for the item scaled by the maximum value of the same item across all firms in the dataset. We note these as “to be standardized” in the following list. Since each third-level index evaluate a specific aspect, we first take the mean of the items under the same third-level index (second-level if there is no third-level), and then we take the mean over the third-level index as our measure of IFCR. For example, we will first take the average of IC11101, IC11102, and IC11103 as the score for IC111. Then ICFR is calculated using the mean of scores of all the third-level (second-level if there is no third-level) index items. As an alternative measure, we also calculate ICFR as the simple mean of the 64 fourthlevel items. Our results remain unchanged using this alternative measure. ICFR Component (fourth level items) IC1: Control Environment IC111: Institutional Arrangement IC11101: Existence of a manual or guideline on internal control in the company. IC11102: Use of an outside professional to help the company improve its internal control. IC113: Board of Directors IC11303: A board-level committee charged with the oversight of internal control (audit committee or risk management committee). IC11304: Number of audit committee members IC114: Board of Supervisors IC11404: Supervisors with legal and accounting expertise. IC121: Internal Control Implementation IC12101:Existence of an internal control department in the company. IC12102:The internal audit department reports to the board. IC131: Recruiting and Training IC13102: The human resource policy contains provisions on recruiting. IC13103: The human resource policy contains provisions on training. 41 Value Example Mean 1 for yes, 0 for no 1.00 0.954 1 for yes, 0 for no 1.00 0.056 1 for yes, 0 for no 1.00 0.870 To be standardized 3.00 0.380 1 for yes, 0 for no 0.00 0.094 1 for yes, 0 for no 1.00 0.817 1 for yes, 0 for no 1.00 0.681 1 for yes, 0 for no 1.00 0.379 1 for yes, 0 for no 1.00 0.421 IC13104: The human resource policy contains provisions on job rotations at critical levels. IC132: Incentives IC13202: The human resource policy contains provisions on evaluations. IC13205: Employee salary is linked to performance. IC133: Severance IC13301: The human resource policy contains provisions on discharging employees. IC13302: The human resource policy contains provisions on employee resignations. IC13303: An audit is performed upon the departure of the Chairman, CEO, and other executives. IC142: Competence variable mean IC14201: Percentage of the staff with at least a junior college education. IC14202: Existence of employee training by the company. IC2: Risk Assessment IC221: Risk evaluation IC22101: Existence of a risk management committee or department. IC222:Internal risk IC22202: A human resource risk disclosed in public filings. IC22203: An operation risk disclosed in public filings. IC22205: A financial risk disclosed in public filings. IC231: Risk Evaluation Method variable mean IC23101: A quantitative risk analysis in the annual report or in the internal control evaluation report. IC23102: Economic risk disclosed? IC241: Response Strategies IC24101: Any response strategy disclosed? IC24102: Any analysis of risk tolerance? IC242: Risk Management Measure IC24201: A risk management measure taken to control risk. 42 1 for yes, 0 for no 0.00 0.079 1 for yes, 0 for no 1.00 0.831 1 for yes, 0 for no 1.00 0.837 1 for yes, 0 for no 0.00 0.124 1 for yes, 0 for no 0.00 0.062 1 for yes, 0 for no 0.00 0.123 To be standardized 0.66 0.395 1 for yes, 0 for no 1.00 0.609 1 for yes, 0 for no 1.00 0.064 1 for yes, 0 for no 0.00 0.112 1 for yes, 0 for no 1.00 0.303 1 for yes, 0 for no 0.00 0.465 1 for yes, 0 for no 0.00 0.032 1 for yes, 0 for no 0.00 0.078 1 for yes, 0 for no 1 for yes, 0 for no 1.00 1.00 0.721 0.524 1 for yes, 0 for no 1.00 0.228 IC3: Control Activities IC31: Separation of Duties IC31101: Controls for incompatible separation of duties. IC31102: Controls for authorizing and approving. IC31103: Approval of material matters is in accordance with existing procedures. IC32: Accounting Control IC32101: Company has an accounting information system. IC34: Budget Control variable IC34101: Existence of a budgeting committee or department. IC34102: A budget is implemented in the company. IC34103: The annual budget is discussed in the shareholder meeting or other similar setting. IC35: Operating Control IC35101:Existence of an operational analysis. IC36: Performance Control IC36101:Existence of a performance analysis committee or department. IC36102:A report is issued from the performance analysis. IC37: Emergency Control IC37101:Existence of a system that generates early warnings for material risk. IC37102:Existence of an emergency response system. IC4: Information and Communication IC41: Information Collection IC41101:Existence of a channel for internal communication of information (e.g. financial analysis and staff meeting) IC41102:Channel for external collection of information (e.g. information from regulatory bodies). IC42: Information Communication IC42201: A system/mechanism governing information disclosure. IC42205: Another platform for communicating with investors. 43 1 for yes, 0 for no 1.00 0.485 1 for yes, 0 for no 1.00 0.670 1 for yes, 0 for no 1.00 0.653 1 for yes, 0 for no 1.00 0.385 1 for yes, 0 for no 1.00 0.058 1 for yes, 0 for no 1.00 0.829 1 for yes, 0 for no 1.00 0.267 1 for yes, 0 for no 1.00 0.996 1 for yes, 0 for no 1.00 0.964 1 for yes, 0 for no 1.00 0.870 1 for yes, 0 for no 1.00 0.109 1 for yes, 0 for no 1.00 0.206 1 for yes, 0 for no 1.00 0.994 1 for yes, 0 for no 1.00 0.995 1 for yes, 0 for no 1.00 0.983 1 for yes, 0 for no 1.00 0.398 IC425: Internal Timeliness IC42501:Periodic reports are released on the scheduled date. IC43: Information System IC43101:Existence of an information department or information security department. IC441: Anti-Fraud Mechanism IC44101:Anti-fraud mechanisms in the company. IC44102:A channel for whistle blowing. IC442: Anti-Fraud Priority IC44201:The company specifies the priorities for anti-fraud. IC5: Monitoring IC51: Internal Monitoring Function IC511: Monitoring from the Internal Audit Department IC51101:Inspection of internal control by the internal audit department. IC512: Monitoring from the Board of Supervisors IC51201:The board of supervisors monitors operational activities, such as financing activities, asset sales, and acquisitions. IC513: Monitoring from the Board of Directors IC51301:The audit committee discusses the internal control inspection in its responsibility report. IC51302:The independent directors discuss the internal control inspection in their responsibility report. IC514: Special Monitoring IC51401:Monitoring for suspicious special events. IC52: Internal Control Deficiencies IC52101:A standard for deficiency recognition. IC52102:An analysis of the reasons for deficiencies. IC52103:A plan for rectifying internal control deficiencies. IC52104:The company tracks the reform of internal control procedures. 44 1 for yes, 0 for no 1.00 0.918 1 for yes, 0 for no 1.00 0.124 1 for yes, 0 for no 0.00 0.068 1 for yes, 0 for no 0.00 0.079 1 for yes, 0 for no 0.00 0.035 1 for yes, 0 for no 1.00 0.688 1 for yes, 0 for no 1.00 0.982 1 for yes, 0 for no 0.00 0.317 1 for yes, 0 for no 0.00 0.269 1 for yes, 0 for no 1.00 0.427 1 for yes, 0 for no 0.00 0.052 1 for yes, 0 for no 0.00 0.305 1 for yes, 0 for no 0.00 0.523 1 for yes, 0 for no 0.00 0.582 IC53: Internal Control Disclosure IC53102:The internal controls are well designed. IC53103:The internal controls function well. IC53104:Performance of an inspection of the internal control system. IC52106:The inspection of the internal control system was evaluated. IC53107:A plan to improve internal control exists. IC53108:A plan for next year’s internal control exists. IC53110:The board of supervisors evaluates the internal control system. IC53111:The independent directors evaluate the internal control system. 45 1 for yes, 0 for no 1.00 0.525 1 for yes, 0 for no 1 for yes, 0 for no 1.00 1.00 0.525 0.443 1 for yes, 0 for no 0.00 0.317 1 for yes, 0 for no 1.00 0.361 1 for yes, 0 for no 0.00 0.084 1 for yes, 0 for no 1.00 0.682 1 for yes, 0 for no 0.00 0.395 Appendix C: Variable Definitions INTERNAL CONTROL MEASURES A measure of the strength of internal control over financial reporting in ICFR year t based on 64 of the 144 items collected by Chen et.al. (2013); see Appendices A&B for details. DEPENDENT VARIABLES Evidence of tunneling (removal of cash from the firm), measured as OTHREC other receivables scaled by year-end total assets; Evidence of tunneling (removal of cash from the firm), measured as the RELATEDLOAN amount of loans from the listed firm to related parties disclosed in the notes to the financial statements, scaled by year-end total assets; Private consumption expenses scaled by sales, including specific PRIVATE expenditures on office supplies, business travel, entertainment, telecommunication, training abroad, board meetings, automobile, conferences, and other expense. Data on these expenditures are collected from note disclosures. An indicator variable equal to one if there is ex post public disclosure of EXPOSED corrupt self-serving behavior by management (e.g. receipt of bribes) in year t (where year t is the year management undertook the corrupt behavior, not necessarily the year it was revealed), zero otherwise; Evidence of earnings management, measured using the absolute value ABSDA of discretionary accruals in year t based on the performance-adjusted cross-sectional modified Jones model (Kothari et al. 2005); Evidence of earnings management, measured as an indicator variable RESTATE equal to one if there is a restatement of year t (discovered in subsequent years), zero otherwise; Operating income divided by average total assets; OI CONTROL VARIABLES ASSETS The logarithm of total assets (in Chinese Yuan); Total sales in year t divided by average total assets. ATO An indicator variable, equal to one if the company is audited by one of BIG4 the four largest U.S. audit firms, zero otherwise; The average of five components related to characteristics of the board BOARD of directors collected by Chen et al. (2013): board size, number of board committees, percentage of independent directors, average attendance ratio of board meetings for independent directors, and the number of dissenting proposals by independent directors. Each component is standardized to range from zero to one (i.e., by scaling by the maximum value of that component) before averaging. An indicator variable equal to one if the CEO is also the Chairman of CEOCHAIR the board, zero otherwise; Percentage change of the amount of total debt; DISSUE Percentage change of the number of outstanding shares; EISSUE An indicator variable equal to one if the company is traded on the EXCHANGE Shenzhen Stock Exchange, zero if the company is traded on the Shanghai Stock Exchange; An indicator variable equal to one if the company exports merchandise EXPORT to foreign countries and districts (including Hong Kong and Taiwan), zero otherwise; 46 FOREIGHSHR Percentage of foreign institutional ownership; The average of four components related to shareholder composition collected by Chen et al. (2013): Chairman of the board is also the controlling shareholder (1 for no, 0 for yes), the ownership percentage of institutional investors, the number of institutional investors in list of top ten shareholders, and the average percentage of shareholders that attend annual shareholder meetings. Each component is standardized to range from zero to one (by scaling each value by the maximum value of that component) before averaging. An indicator variable equal to one if the company lists on the Shenzhen LIST2006 or Shanghai Exchanges after 2005, zero if they were already listed; An indicator variable equal to one if the company reports a loss in the LOSS year, zero otherwise; An indicator variable equal to one if the company completed a merger M&A or acquisition during the year, zero otherwise; MARKETIZATION A comprehensive index measuring the development of the regional market in which a firm is registered based on Fan and Wang (2006) and Jiang et al. (2010); Market value of equity divided by book value of equity. MB Percentage of outstanding shares held by top executives (multiplied by MGMTSHR 100); Ratio of the average executive total compensation relative to other PAYRATIO employees’ average total compensation; Earnings before extraordinary items divided by average total assets; ROA The logarithm of total sales (in Chinese Yuan); SALES The logarithm of the number of the geographic segments; SEGMENTS An indicator variable equal to one if the company is controlled by the SOE central or local government or their agencies, zero otherwise; The percentage of outstanding shares held by the largest shareholder; TOP1SHR The sum of the ownership percentages of the second to the fifth largest TOP2to5 shareholders; An indicator variable, equal to one if the sum of the ownership TOP2to5IND percentages of the second to the fifth largest shareholders is above or equal to the median,and zero otherwise. INVCOMP 47 Appendix D: Determinants and validation of ICFR Score In this appendix we first explore the determinants of our ICFR Score to assess whether it exhibits similar determinants as previously examined measures of ICFR strength (Section D1). We next examine the association between the internal control score and earnings quality as a validity test of our internal control measure (Section D2). D.1 Determinants of ICFR Score We develop our internal control prediction model based on determinants considered in the U.S. literature as well as factors specific to the China market. Specifically, we estimate the following first stage regression: ICFR = β0 + β1SOE +β2ASSETS + β3ROA + β4 LOSS + β5 MB + β6 EXPORT + β7 SEGMENTS + β8 M&A + β9 FOREIGNSHR + β10BIG4+ β11LIST2006 (D1) + β12 MARKETIZATION + β13 EXCHANGE + ε Our first determinant is SOE, which equals one if the controlling shareholder of a firm is either a central or local government or their agencies. The inclusion of this variable is based on our lengthy discussion above regarding the governmental pressure on SOE firms to establish internal controls. We next follow prior literature and include the following variables that have been shown to determine a company’s internal control strength. Doyle et al. (2007b) find that internal control problems are more prevalent in firms that are smaller, financially weaker, growing more quickly, and more complex. They argue that larger firms and more profitable firms have more resources to invest in internal control, thus we include firm size (ASSETS), and return on assets (ROA), and expect the strength of internal controls to be increasing with both ASSETS and ROA. Similarly, firms going through financial distress are less likely to invest resources in internal control. We include LOSS that equals one if the company reports a loss in the year to proxy for poor financial health. We include the ratio of the market value of equity to book value of equity (MB) to proxy for growth. We measure financial reporting complexity by the existence of exporting transactions (EXPORT) and the number of segments (SEGMENTS). Ashbaugh-Skaife et al. (2008) find that structural changes within a company could affect the internal control strength, thus we include an indicator variable M&A that equals one if a firm completes a merger or acquisition during the year. As suggested in Ashbaugh-Skaife et al. (2007), shareholder composition and auditors may also influence a firm’s choice in its internal control. We include the percentage of shares held by foreigners and expect that a greater concentration of foreigners would reflect greater demand for strong oversight and thus internal controls. We include an indicator variable equal to one if the company is audited by one of the four biggest U.S. accounting firms to proxy for auditor oversight. In addition to the determinants established in prior research in the U.S., we include the following variables specific to Chinese firms’ internal control strength. First, 2006 was the year China moved to advance corporate internal control. In 2006 both the Shenzhen and Shanghai exchanges issued several regulations regarding internal control for firms going 48 public. 35 As a result, any firm going public in 2006 or afterwards is subject to higher regulatory standards for internal controls. Therefore, we include an indicator variable LIST2006 that takes a value of one for firms that listed on the Shenzhen and Shanghai exchanges on or after 2006. Second, we recognize the difference in regulations on internal controls by the two Chinese stock exchanges in our sample (Shanghai vs. Shenzhen) by including an indicator EXCHANGE that equals one if a firm is listed on the Shenzhen exchange, and zero otherwise. Finally, we include MARKETIZATION, which measures the development of the regional market in which the firm is registered (Fan and Wang 2006, Jiang et al. 2010). We also include year and industry fixed effects. The results are reported in Table D1. The results in Table D1 suggest that state-owned enterprises (SOE) tend to have stronger internal controls; the coefficient on SOE is significantly positive (coefficient=0.010; p-value=0.010), consistent with SOEs experiencing government pressure to establish internal controls. Consistent with prior literature using U.S. internal control disclosures, we find that larger and more profitable firms have higher internal control scores. The coefficient on firm size (ASSET) is significantly positive. As expected, there is a positive relation between ROA and ICFR, and the coefficient on LOSS is significantly negative (i.e., loss firms tend to have weaker internal controls). Unlike studies using U.S. data, we find that more complex firms (SEGMENTS) tend to have stronger internal controls. It is important to point out that U.S. studies use the ex post existence of ineffective internal controls to identify internal control strength, and complexity likely leads to the existence of underlying problems. Our measure, in contrast, examines the policies and procedures in place, and thus likely reflects that more complex firms need, and thus choose to implement, stronger internal controls. We find no evidence that the composition of foreign shareholders is associated with internal control strength, but we see that companies audited by the Big Four U.S. audit firms tend to maintain stronger internal controls. Consistent with the higher internal control requirements for firms that listed on or after 2006, and firms listed on the Shenzhen Stock Exchange, we find positive coefficients on LIST2006 and EXCHANGE. Finally, we find a significantly positive association between internal control over financial reporting and the development of regional markets (MARKETIZATION). We also estimate Equation (D1) separately within the SOE and non-SOE samples (excluding SOE as a main effect). We find that the adjusted R-square is lower within SOEs (27.5%) than non-SOEs (33.2%). The lower explanatory power among SOEs is consistent with our broader conclusion that ICFR within SOEs may be driven by governmental pressure incremental to the underlying economic determinants identified in prior research. D.2: Validating the Measure of ICFR Strength To the extent that our measure of internal control strength captures the strength of internal control, we expect to observe a positive association between ICFR and earnings quality. 35 For example, in the IPO process, a new regulation requires a firm going public to disclose the management’s assessment as well as the auditor’s opinion on its internal control effectiveness. 49 We examine two measures of earnings quality: the absolute value of discretionary accruals (ABSDA) and the occurrence of accounting errors (RESTATE). Earnings Quality = β0 +β1ICFR +β2ASSETS+β3OCF +β4LEV +β5ROA +β6MB (D2) +β7EISSUE +β8DISSUE +β9BIG4 +β10TOP1SHR+ ε The dependent variable is ABSDA or RESTATE. ABSDA is the absolute value of the residuals from the performance-adjusted cross-sectional modified Jones model (Kothari et al. 2005). RESTATE equals one if the firm subsequently restated an accounting error made in the current year, and zero otherwise. We control for variables documented in prior literature that influence earnings quality including cash flows from operations (OCF), leverage (LEV), firm accounting performance measured by returns on assets (ROA), and growth opportunities measured by market-to-book ratio (MB). We control for new issuances of equity and debt (EISSUE and DISSUE) because firms have stronger incentives to manage earnings in the presence of financing needs (Dechow et al. 1996). We also include audit quality, which is an important factor in determining earnings quality; BIG4 is equal to one if the auditor is one of the four largest U.S. accounting firms, and zero otherwise. Lastly, we include the percentage of shares owned by the largest shareholder (TOP1SHR) to control for potential effects of ownership structure on accounting quality (Dechow et al. 1996). Finally we include year and industry fixed effects. Overall, we find that ICFR is negatively associated with both the absolute value of discretionary accruals and the occurrence of accounting errors, suggesting that earnings quality is lower for firms with weaker internal controls. The coefficients on the control variables are largely consistent with prior literature. We also examine the association between ICFR and earnings quality for the SOE firms and the non-SOE firms separately. When examining discretionary accruals, the coefficient appears slightly larger in magnitude among non-SOEs, but the significance level weaks to extremely marginal. When examining restatements we find that, for both SOE and non-SOE firms, ICFR has a significant negative association. Again, the magnitude of the coefficient on ICFR for non-SOEs appears larger than that of SOEs, but the difference is not statistically significant. Taken together, the findings in Appendix D provide support for the validity of our internal control strength measure. 50 Table D1: Determinants of Internal Control over Financial Reporting Strength Intercept Predicted Signs ? SOE + ASSETS + ROA + LOSS – MB – EXPORT – SEGMENTS ? M&A – FOREIGHSHR + BIG4 + LIST2006 + MARKETIZATION + EXCHANGE ? Year effects Industry effects No. of Obs. Adjusted R2 Full Sample SOEs Non-SOEs -0.118*** (0.004) 0.010** (0.010) 0.022*** (0.000) 0.038* (0.068) -0.018*** (0.001) 0.0003 (0.829) 0.003 (0.794) 0.013*** (0.001) 0.004 (0.880) -0.007 (0.600) 0.012* (0.087) 0.043*** (0.000) 0.002** (0.043) 0.041*** (0.000) Included Included -0.132** (0.013) -0.163** (0.012) 0.021*** (0.000) 0.047 (0.111) -0.009 (0.106) 0.0001 (0.544) 0.004 (0.772) 0.010** (0.026) 0.002 (0.693) 0.030 (0.183) 0.008 (0.202) 0.039*** (0.000) 0.004*** (0.001) 0.036*** (0.000) Included Included 0.024*** (0.000) 0.013 (0.347) -0.032*** (0.000) 0.0004 (0.881) 0.000 (0.526) 0.017** (0.012) 0.008 (0.913) -0.075 (0.953) 0.013 (0.305) 0.054*** (0.000) -0.003 (0.944) 0.052*** (0.000) Included Included 4,775 0.298 3,196 0.275 1,579 0.332 All variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. Pvalues reported in parentheses are based on one-tailed tests for variables with predicted signs and twotailed tests for variables without predicted signs. Standard errors are clustered by firm. 51 Table D2: Internal Control and Earnings Quality Dependent Variable Predicted Full Sample Signs Intercept ? 0.321*** (0.000) ICFR -0.034** (0.035) ASSETS -0.013*** (0.000) OCF -0.110*** (0.002) LEV + 0.043*** (0.000) ROA ? 0.100** (0.015) MB + 0.0002 (0.276) EISSUE + 0.047*** (0.000) DISSUE + 0.043*** (0.000) BIG4 0.016 (0.993) Top1SHR -0.004 (0.372) Year effects Included Industry effects Included No. of Obs. Adjusted (or Pseudo) R2 4,735 0.230 ABSDA (OLS) RESTATE (Logit) SOEs Non-SOEs Full Sample SOEs Non-SOEs 0.300*** (0.000) -0.034* (0.061) -0.012*** (0.000) -0.082** (0.005) 0.056*** (0.000) 0.042 (0.343) 0.0004 (0.221) 0.055*** (0.000) 0.041*** (0.000) 0.016 (0.996) 0.006 (0.656) Included Included 0.475*** (0.000) -0.046 (0.108) -0.018*** (0.000) -0.147*** (0.010) 0.032*** (0.001) 0.177*** (0.009) -0.0003 (0.648) 0.038*** (0.007) 0.047*** (0.000) 0.019 (0.786) -0.024 (0.212) Included Included 0.420 (0.780) -1.409** (0.019) -0.088 (0.108) -0.878 (0.120) 0.076 (0.321) -1.704** (0.035) 0.012 (0.138) 0.175 (0.231) -0.027 (0.611) -1.132** (0.021) -0.987** (0.017) Included Included 1.128 (0.577) -1.141* (0.088) -0.131* (0.088) -0.607 (0.268) 0.152 (0.310) -2.530** (0.030) -0.006 (0.630) 0.540** (0.025) -0.040 (0.632) -1.531*** (0.006) -0.438 (0.211) Included Included 0.269 (0.915) -1.978** (0.041) -0.054 (0.319) -0.958 (0.208) 0.006 (0.488) -0.303 (0.793) 0.029** (0.019) -0.356 (0.775) 0.022 (0.444) 0.034 (0.514) -3.172*** (0.001) Included Included 3,176 0.230 1,559 0.216 4,760 0.055 3,185 1,575 0.065 0.068 OCF is operating cash flows divided by end-of-year total assets; LEV is leverage calculated by total liabilities divided by total assets; All other variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. P-values reported in parentheses are based on two-tailed tests. Standard errors are clustered by firm. 52 Table 1 Panel A: Sample Selection Firm-years listed on Shenzhen and Shanghai exchanges 2007-2010* Firm-year observations 6,764 Less:Firms in the financial industry Firms traded on small and medium-sized enterprises(SMEs) board Firms with missing data required for determinants of internal control strength Total sample used in multivariate analysis (120) (1,401) (468) 4,775 *Firms traded on the Entrepreneur Section of the Shenzhen Exchange are excluded because the Entrepreneur Section was not established until 2009 and these firms are not covered by the Internal Control Index constructed by Chen et al. (2013). Panel B: Distribution by Year Year 2007 2008 2009 2010 Total SOEs N 787 802 802 805 3,196 Non-SOEs % 66.41 67.28 66.78 67.25 66.93 53 N 398 390 399 392 1,579 % 33.59 32.72 33.22 32.75 33.07 Total 1,185 1,192 1,201 1,197 4,775 Table 2, Panel A: Descriptive Statistics for Full Sample Variable N Mean Median SD Q1 Q3 ICFR Score 4,775 0.465 0.460 0.108 0.388 0.535 OTHREC 4,775 0.027 0.012 0.042 0.004 0.030 RELATEDLOAN 4,775 0.012 0.000 0.042 0.000 0.003 PRIVATE 2,128 0.027 0.010 0.056 0.004 0.024 EXPOSED 2,796 0.045 0.000 0.207 0.000 0.000 Total Assets 4,775 6,965.7 2,504.1 15,204.4 1,142.3 5,693.1 Sales Revenue 4,775 4,934.8 1,443.7 11,239.0 595.3 3,573.8 # of Segment 4,775 3.486 2.000 2.840 2.000 5.000 ROA 4,775 0.036 0.033 0.080 0.011 0.065 LOSS 4,775 0.122 0.000 0.328 0.000 0.000 MB 4,775 4.342 3.387 5.379 1.980 5.496 EXPORT 4,775 0.414 0.000 0.493 0.000 1.000 M&A 4,775 0.780 1.000 0.414 1.000 1.000 FOREIGNSHR 4,775 0.019 0.000 0.073 0.000 0.000 BIG4 4,775 0.065 0.000 0.246 0.000 0.000 LIST2006 4,775 0.059 0.000 0.236 0.000 0.000 MARKETIZATION 4,775 8.737 8.660 2.043 7.190 11.040 EXCHENGE 4,775 0.377 0.000 0.485 0.000 1.000 TOP1SHR 4,775 0.357 0.337 0.154 0.232 0.476 TOP2to5 4,775 0.104 0.075 0.088 0.034 0.159 INVCOMP 4,775 0.304 0.302 0.163 0.176 0.409 BOARD 4,775 0.512 0.512 0.060 0.490 0.541 CEOCHAIR 4,722 0.132 0.000 0.338 0.000 0.000 PAYRATIO 3,954 3.734 2.763 2.940 1.822 4.592 MGMTSHR 4,773 0.530 0.002 3.258 0.000 0.017 ATO 4,775 0.746 0.625 0.546 0.364 0.963 EISSUE 4,762 0.117 0.000 0.280 0.000 0.009 DISSUE 4,773 0.241 0.107 0.659 -0.048 0.331 Total Assets and Sales Revenues are reported in millions of Chinese Yuan. 54 Table 2, Panel B: Descriptive Statistics for the SOE and Non-SOE Samples SOEs Non-SOEs N=3,196 N=1,579 Test of Differences Mean Median Mean Median Mean Test Median Test ICFR Score 0.475 0.468 0.446 0.441 <0.001 <0.001 OTHREC 0.022 0.011 0.038 0.017 <0.001 <0.001 RELATEDLOAN 0.009 0.000 0.016 0.000 <0.001 <0.001 PRIVATE 0.020 0.009 0.040 0.014 <0.001 <0.001 EXPOSED 0.062 0.000 0.011 0.000 <0.001 <0.001 Total Assets 8,867.4 2,958.6 3,116.4 1,530.6 <0.001 <0.001 Sales Revenue 6,297.7 1,812.6 2,176.0 882.1 <0.001 0.001 # of Segments 3.422 2.000 3.616 2.000 0.027 <0.001 ROA 0.034 0.032 0.039 0.037 0.038 0.008 LOSS 0.114 0.000 0.140 0.000 0.009 0.009 MB 4.120 3.170 4.791 3.807 <0.001 <0.001 EXPORT 0.431 0.000 0.379 0.000 <0.001 <0.001 M&A 0.755 1.000 0.832 1.000 <0.001 <0.001 FOREIGNSHR 0.020 0.000 0.016 0.000 0.105 0.205 BIG4 0.084 0.000 0.027 0.000 <0.001 <0.001 LIST2006 0.068 0.000 0.042 0.000 <0.001 <0.001 MARKETIZATION 8.686 8.660 8.841 8.810 0.014 0.060 EXCHANGE 0.369 0.000 0.392 0.000 0.121 0.121 TOP1SHR 0.384 0.382 0.302 0.262 <0.001 <0.001 TOP2to5 0.100 0.066 0.112 0.093 <0.001 <0.001 INVCOMP 0.308 0.304 0.297 0.294 0.022 0.047 BOARD 0.512 0.511 0.513 0.513 0.969 0.146 CEOCHAIR 0.095 0.000 0.205 0.000 <0.001 <0.001 PAYRATIO 3.542 2.585 4.126 3.223 <0.001 <0.001 MGMTSHR 0.066 0.002 1.469 0.002 <0.001 0.001 ATO 0.784 0.652 0.668 0.537 <0.001 <0.001 EISSUE 0.107 0.000 0.138 0.000 <0.001 0.004 0.006 <0.001 DISSUE 0.260 0.126 0.204 0.067 Total Assets and Sales Revenues are reported in millions of Chinese Yuan. 55 Table 2, Panel C: Mean ICFR Score and Corruption Measures by Industry RELATED Industry N* ICFR OTHREC PRIVATE LOAN Agriculture, forestry 119 0.427 0.040 0.018 0.022 and fishing Mining EXPOSED 0.017 105 0.505 0.013 0.010 0.008 0.219 2,725 0.461 0.023 0.009 0.026 0.033 Utilities 242 0.478 0.014 0.005 0.013 0.095 Construction 104 0.516 0.053 0.007 0.010 0.077 Transportation and warehousing 197 0.499 0.021 0.010 0.025 0.152 Information technology 270 0.454 0.045 0.019 0.046 0.037 Distribution and retail 305 0.462 0.034 0.020 0.024 0.023 Real estate 290 0.482 0.029 0.011 0.028 0.041 Service 132 0.482 0.033 0.006 0.026 0.023 Communication and mass media 41 0.444 0.036 0.010 0.093 0.000 Other Industries 245 0.441 0.048 0.026 0.057 0.029 Manufacturing Total 4,775 0.465 0.027 0.012 0.027 0.045 *The number of observations varies for each variable due to differences in data availability. N refers to the number of observations corresponding to ICFR. Table 2, Panel D: Mean ICFR Score and Corruption Measures by Year Year ICFR OTHREC RELATEDLOAN PRIVATE EXPOSED 2007 0.408 0.035 0.016 0.032 0.047 2008 0.446 0.029 0.013 0.027 0.049 2009 0.489 0.024 0.009 0.027 0.047 2010 0.517 0.022 0.008 0.024 0.036 Total 0.465 0.027 0.012 0.027 0.045 56 Table 2, Panel E: Pearson (Spearman) Correlation a a 0.01 -0.07 a a a -0.05 a -0.04 a 0.05 -0.08 b 0.02 -0.05 a PRIVATE -0.11 0.27 0.10 1.00 -0.03 0.08 0.06 -0.31 -0.24 -0.06 -0.03 0.09 -0.08 -0.06 -0.03 EXPOSED 0.05 a 0.00 0.01 -0.07 a 1.00 -0.02 -0.02 0.13 a 0.06 a 0.06 a -0.01 -0.01 0.02 -0.02 -0.03 ABSDA -0.05 a 0.09 a -0.02 1.00 RESTATE -0.12 a 0.08 b -0.02 0.04 ASSETS 0.34 a -0.25 a 0.09 a -0.10 ATO 0.09 a -0.08 a 0.06 a -0.06 BIG4 0.15 a -0.08 a .058 a -0.06 b 0.02 0.06 a 0.02 a 0.06 0.05 a 0.00 -0.39 a -0.04 a -0.35 a 0.09 a -0.10 a a 0.05 BOARD 0.28 CEOCHAIR -0.06 a -0.06 -0.04 a -0.04 a -0.12 a a 0.04 a -0.07 DISSUE 0.14 EISSUE 0.10 EXCHANGE 0.15 EXPORT 0.06 0.02 0.00 -0.11 0.06 a -0.06 INVCOM -0.02 -0.06 LIST2006 0.05 a -0.08 LOSS -0.13 a 0.10 M&A 0.01 0.13 a -0.06 a 0.07 -0.03 0.04 a -0.02 -0.08 a a 0.06 a a a 0.07 a a 0.11 0.04 -0.08 a -0.05 0.10 a -0.04 b a -0.16 a a 0.12 a -0.16 a -0.23 ROA 0.13 SALES 0.32 SEGMENTS 0.09 SOE 0.12 TOP1SHR 0.09 0.05 0.01 0.09 a -0.09 1.00 0.13 a 0.40 a 0.07 a -0.14 a 0.18 a -0.10 a 0.26 a 0.19 a 0.15 a -0.20 a 0.04 a 0.05 a 0.03 -0.12 a -0.40 a 0.07 a 0.01 a 0.40 a 0.14 a 0.13 a -0.03 b -0.04 a 0.13 a a -0.01 -0.01 0.00 a 0.00 -0.03 0.12 a -0.05 a 0.32 a a -0.02 a -0.04 1.00 0.00 b 0.04 0.01 1.00 -0.02 -0.03 a 1.00 a 0.33 -0.03 b 0.18 a 1.00 -0.05 0.07 0.01 -0.14 a -0.07 a -0.05 a 0.01 a a 0.08 a 0.07 a 0.07 -0.05 a 0.05 a -0.01 -0.01 -0.03 a 0.07 -0.02 0.03 -0.04 0.34 a 0.00 0.23 b 0.00 a -0.04 b a a -0.09 -0.03 a 0.23 a a a 0.04 a 0.12 b a -0.04 0.02 a 007 a 0.06 0.04 a 0.36 a -0.06 0.02 0.14 a -0.24 a 0.13 a -0.06 a -0.03 b 0.00 0.07 a 0.01 0.00 -0.05 a b 0.00 a a 0.06 0.14 a 0.07 a -0.14 a -0.04 0.00 -0.05 0.10 b 0.07 a -0.04 0.08 a 0.02 -0.29 b -0.02 -0.03 0.02 a 0.11 0.07 a -0.06 a 0.00 -0.03 0.20 a 0.12 a a 0.05 0.03 a 0.06 a -0.17 a -0.03 b -0.24 0.02 a 0.12 a 0.10 a b -0.08 a -0.08 0.03 a 0.20 a 0.85 0.57 -0.02 0.12 a -0.01 0.27 a -0.05 a 0.30 -0.04 -0.12 a -0.04 1.00 0.15 a 0.18 a 0.00 a 1.00 0.07 a -0.07 a -0.05 a 0.05 a 0.06 a -0.19 a 0.05 a 0.18 a 0.07 a 1.00 -0.04 a 0.04 a -0.08 a -0.12 a 0.04 a -0.02 0.01 -0.07 a -0.04 a -0.01 0.00 0.03 b 0.06 a 0.01 -0.02 -0.03 a -0.01 0.02 -0.01 a a 0.10 a 0.05 a 0.04 b b -0.10 a -0.04 a 0.07 0.09 0.30 0.14 0.12 a 0.13 -0.03 -0.12 0.09 0.03 0.00 0.01 -0.03 a a 0.05 a 0.04 a 0.06 a 0.07 a 0.08 -0.03 0.04 0.05 a 0.03 0.09 a 0.10 a 0.01 -0.02 0.15 a 0.05 a -0.14 a 0.25 a 0.03 a 0.11 a 0.15 a b a 0.02 a 0.21 a 0.18 b -0.01 -0.03 0.03 a -0.02 -0.15 a 0.09 a -0.02 -0.13 a 0.15 a -0.03 a -0.07 b 0.08 a a 0.05 0.03 b 0.02 0.08 a 0.06 a -0.04 a -0.15 a -0.03 -0.15 a -0.13 b -0.02 0.01 0.01 a 0.05 a 0.11 a 0.14 0.00 a 0.04 a 0.19 a -0.04 0.01 0.06 a 0.08 a 0.17 a 0.13 a 0.04 a -0.05 a 0.10 a -0.02 -0.10 a 0.04 b 0.08 b 0.07 a 0.11 a -0.12 a -0.09 b 0.02 0.09 0.16 a 0.00 0.01 a b 0.03 -0.02 a a 0.00 0.04 0.13 b 0.00 -0.02 -0.05 0.05 a -0.01 -0.09 0.00 a 0.00 0.22 a 0.03 a 0.03 a 0.05 b 0.05 b 0.03 a 0.03 a 0.02 0.05 a 0.08 a 0.01 0.23 0.10 a 0.13 a 0.16 -0.04 a -0.01 0.01 -0.02 0.12 b 0.00 0.05 a 0.13 -0.04 a 1.00 -0.01 -0.05 0.01 -0.02 -0.08 a -0.65 a -0.22 -0.02 -0.04 -0.01 -0.01 -0.01 1.00 0.03 0.02 0.01 0.05 a 0.00 -0.03 0.02 -0.09 b 0.03 0.09 a -.001 0.03 1.00 a a a 0.10 0.00 0.01 -0.04 b -0.02 0.03 a -0.14 a -0.03 -0.06 a a 0.04 b 0.00 0.02 -0.20 a 0.04 a 0.21 a 0.12 a -0.09 b 0.03 -0.08 -0.05 a 0.09 a 0.00 -0.09 a a 0.06 -0.57 a 0.10 a -0.21 0.17 a b 0.02 0.18 -0.10 -0.01 -0.02 a a 0.05 0.00 0.00 0.02 0.05 0.07 b 0.14 -005 0.19 0.03 0.03 0.11 a b a 0.01 -0.10 a 0.11 a 0.00 -0.02 -0.01 0.10 b 0.09 0.21 a 3.00 a a 0.04 a -0.04 -0.10 b 0.04 0.04 a 0.02 0.03 a 0.06 a 0.07 a a 0.05 0.03 b 0.00 0.17 a 1.00 0.14 a a 0.15 a 0.06 a 0.19 1.00 0.20 -0.01 -0.03 a -0.21 a 0.11 a 0.22 a 0.24 a 1.00 0.18 a 0.30 a 0.01 a 1.00 a -0.04 a 0.27 a -0.05 1.00 0.25 a 0.15 a 0.29 a 0.01 0.27 a 1.00 0.07 -0.02 0.12 a 0.11 a -0.12 a -0.09 a -0.06 -0.01 0.03 a a a -0.09 a 1.00 a a a 0.05 0.04 -0.02 a 0.07 a -0.04 a a 0.09 0.16 a a -0.03 a 0.01 a 0.00 b 0.05 a 0.09 a a a a 0.05 a 0.12 a -0.13 a -0.09 0.00 -0.04 a -0.03 -0.10 -0.05 a 0.03 b -0.01 -0.23 a 0.16 -0.04 a 1.00 -0.07 b b a -0.05 b b b a a 0.29 -0.01 -0.08 b a 0.05 a a -0.04 a b a a a a b a 0.00 0.33 a b a 0.54 a a -0.02 0.03 a 0.07 a -0.04 0.09 a 0.17 a 0.06 a a a 0.11 0.04 0.02 0.34 0.08 a 0.11 a 0.03 -0.10 a 0.12 -0.04 -0.07 0.02 a b a 0.07 -0.05 -0.07 a 0.05 -0.04 -0.01 0.86 0.05 -0.19 -0.12 -0.02 a a a -0.05 a 0.17 0.05 -0.01 a -0.14 a a a a 0.07 -0.05 0.11 0.21 -0.01 a a -0.02 0.01 0.13 -0.18 0.12 a a -0.05 -0.17 0.02 -0.18 0.01 1.00 -0.02 a a a -0.13 -0.43 -0.09 0.01 b a a a -0.02 1.00 0.13 a a 0.01 b -0.13 0.02 -0.11 -0.06 -0.05 0.01 -0.09 a -0.20 a b a -0.07 a a 0.03 0.06 b -0.08 a 0.09 a -0.03 -0.16 -0.06 0.00 -0.18 a 0.13 0.00 -0.01 a -0.18 a -0.30 a a 0.00 -0.03 a 0.01 b 0.04 a 0.09 a 0.3 -0.02 0.04 a 0.35 a 0.02 a a a 0.11 a -0.03 0.07 b a a a a a a 0.07 0.00 a a a 0.17 a 0.06 0.09 0.20 b -0.01 a a a a b 0.14 a a a a a 0.02 a a -0.12 -0.04 -0.05 0.04 a 0.03 -0.03 -0.05 -0.02 -0.14 0.01 a 0.02 0.01 1.00 -0.22 0.10 0.04 a 0.02 a a a a 1.00 -0.13 a a 0.04 0.09 -0.48 0.22 a a -0.07 a a 0.10 -0.01 0.10 -0.03 0.00 -0.02 -0.06 0.00 -0.02 b 0.14 a a a -0.03 0.08 -0.23 0.04 -0.05 -0.05 a 0.01 a a 0.01 a -0.02 a 0.06 a -0.09 -0.04 a a a 0.10 0.00 0.10 -0.10 -0.03 a 0.04 0.18 0.03 0.05 -0.02 0.04 a 0.20 0.01 0.11 -0.03 0.00 -0.05 a 0.00 0.00 a PAYRATIO -0.03 0.00 -0.05 0.08 0.13 -0.03 0.03 0.03 -0.02 -0.03 0.00 a 0.00 0.03 0.04 -0.06 0.14 a -0.12 0.21 -0.04 -0.04 -0.07 b a -0.02 a a MGMTSHR b b 0.01 a b -0.09 0.06 0.04 0.06 0.01 a -0.08 -0.02 -0.02 b 0.00 a -0.14 -0.02 0.04 -0.03 a a MB a 009 0.03 -0.10 0.04 0.05 -0.06 a 0.09 a 0.01 a 0.03 b a -0.01 -0.05 -0.02 a 0.11 0.10 -0.01 -0.05 a b 0.06 a a b b -0.07 -0.05 0.00 a a a -0.13 b -0.06 a -0.03 -0.01 a -0.06 -0.03 a 0.06 -0.07 a -0.04 0.12 a a 0.11 -0.08 0.01 a -0.05 a a a -0.09 b -0.05 -0.14 a -0.08 -0.01 a -0.07 -0.01 b a 0.06 0.02 a b -0.06 a 1.00 a a a -0.03 a a a a FOREIGHSHR MARKET 0.06 a b a a 0.08 TOP1SHR a -0.08 b -0.04 -0.01 a SOE a 0.05 -0.04 -0.09 a a 0.06 a SEGMENTS a -0.05 a 0.13 a SALES a -0.12 a a 0.06 a ROA -0.17 a 0.06 PAYRATIO a a a MGMTSHR 0.04 0.14 a MB a 0.19 1.00 -0.06 a MARKET 0.10 0.37 a 0.23 a M&A -0.02 a a 0.16 a LOSS -0.27 0.36 -0.05 a 0.08 a LIST2006 0.06 a -0.12 a 0.37 INVCOM 0.13 a FOREIGHSHR -0.12 EXPORT b a EXCHANGE -0.04 0.00 EISSUE 0.06 DISSUE a CEOCHAIR -0.14 BOARD a a RELATEDLOAN BIG4 -0.12 1.00 a a ATO -0.15 a 1.00 OTHREC a ASSETS RESTATE ABSDA EXPOSED a PRIVATE RELATEDLOAN OTHREC ICFR ICFR b a 0.01 a -0.11 a -0.10 b a 0.14 a 0.32 -0.05 a a a a a 0.01 a All variables are described in Appendix. Pearson correlations are reported above the diagonal, and Spearman correlations are reported in the off diagonal. All continuous variables are winsorized at 1% and 99% to mitigate outliers. a, b, indicate statistical significance at the 0.01, 0.05 level, respectively. 57 Table 3: Internal Control and Tunneling Panel A: Full Sample Intercept Predicted Signs ? ICFR − ASSETS − ROA − TOP1SHR − INVCOMP − BOARD − CEOCHAIR + MARKETIZATION − BIG4 − EXCHANGE ? Year effects Industry effects Dep. Var.= OTHREC 0.183*** (0.000) -0.013** (0.048) -0.006*** (0.000) -0.045*** (0.001) -0.027*** (0.000) -0.012** (0.012) 0.011 (0.836) 0.001 (0.373) -0.001*** (0.005) 0.006 (0.994) 0.002 (0.276) Included Included Dep. Var.= RELATEDLOAN 0.132*** (0.000) -0.021*** (0.010) -0.005*** (0.000) -0.036** (0.015) -0.012*** (0.009) -0.002 (0.358) -0.004 (0.349) -0.003 (0.574) 0.001 (0.893) 0.013 (1.000) 0.001 (0.515) Included Included 4,722 0.139 4,722 0.053 No. of Obs. Adjusted R2 All variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. P-values reported in parentheses are based on one-tailed tests for variables with predicted signs and two-tailed tests for variables without predicted signs. Standard errors are clustered by firm. 58 Table 3: Internal Control and Tunneling (continued) Panel B: SOEs versus Non-SOEs Dep. Var. OTHREC Predicted SOEs Non-SOEs Signs ? 0.144*** 0.229*** Intercept (0.000) (0.000) − -0.006 -0.032** ICFR (0.207) (0.025) − -0.004*** -0.008*** ASSETS (0.000) (0.000) − -0.074*** -0.021 ROA (0.000) (0.178) − -0.021*** -0.025** TOP1SHR (0.001) (0.014) − -0.006 -0.026*** INVCOMP (0.156) (0.007) − -0.006 0.060 BOARD (0.250) (0.929) + 0.004 -0.003 CEOCHAIR (0.129) (0.779) − -0.0002 -0.003*** MARKETIZATION (0.307) (0.001) − 0.002 -0.004 BIG4 (0.845) (0.165) ? -0.001 0.007* EXCHANGE (0.744) (0.076) Included Included Year effects Included Included Industry effects No. of Obs. Adjusted R2 3,154 0.131 1,568 0.148 RELATEDLOAN SOEs Non-SOEs 0.089*** (0.000) -0.006 (0.240) -0.003*** (0.000) -0.066*** (0.001) -0.009** (0.037) -0.002 (0.331) -0.014* (0.086) 0.001 (0.187) 0.001 (0.962) 0.010 (0.998) 0.003* (0.080) Included Included 0.204*** (0.001) -0.049*** (0.006) -0.008*** (0.000) -0.0004 (0.494) -0.016* (0.068) 0.00002 (0.501) 0.031 (0.753) -0.006 (0.528) 0.0004 (0.665) 0.008 (0.936) -0.003 (0.519) Included Included 3,154 0.049 1,568 0.070 Test of differences between coefficients for OTHREC SOEs vs. non-SOEs Coefficient Diff. Z-statistic ICFR 0.025* 1.409 p-value 0.068 Test of differences between coefficients for RELATEDLOAN SOEs vs. non-SOEs Coefficient Diff. Z-statistic 0.043** 2.035 ICFR p-value 0.021 All variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. P-values reported in parentheses are based on one-tailed tests for variables with predicted signs and two-tailed tests for variables without predicted signs. Standard errors are clustered by firm. 59 Table 4: Internal Control and Private Consumption Dependent variable= PRIVATE Predicted Signs Intercept ? ICFR − PAYRATIO − MGMTSHR − ATO − ROA ? LOSS ? EISSUE + DISSUE + BIG4 − TOP1SHR − INVCOMP − BOARD − CEOCHAIR + MARKETIZATION − Year effects Industry effects No. of Obs. Adjusted R2 Full Sample SOEs Non-SOEs 0.071*** (0.006) -0.025** (0.047) -0.001** (0.015) 0.001 (0.810) -0.018*** (0.000) 0.0003 (0.990) 0.019*** (0.003) -0.002 (0.731) -0.001 (0.707) -0.002 (0.245) -0.026*** (0.004) -0.005 (0.268) -0.026 (0.223) 0.006 (0.136) -0.0004 (0.352) Included Included 0.027 (0.122) -0.002 (0.459) -0.000 (0.354) -0.002** (0.010) -0.015*** (0.000) -0.010 (0.724) 0.009 (0.168) -0.0003 (0.546) -0.001 (0.663) -0.008*** (0.004) -0.012 (0.104) -0.010 (0.133) 0.013 (0.708) 0.003 (0.301) 0.001 (0.878) Included Included 0.195*** (0.004) -0.069* (0.027) -0.003*** (0.004) 0.001 (0.837) -0.026*** (0.000) 0.052 (0.282) 0.034*** (0.009) -0.002 (0.611) 0.0001 (0.493) 0.009 (0.731) -0.057** (0.011) -0.001 (0.485) -0.124* (0.082) 0.009 (0.273) -0.004* (0.098) Included Included 1,796 0.111 1,235 0.068 561 0.162 Test of coefficient difference ICFR Coefficient Diff. 0.068** SOE vs. Non-SOE Z-statistic 1.744 p-value 0.040 All variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. Pvalues reported in parentheses are based on one-tailed tests for variables with predicted signs and twotailed tests for variables without predicted signs. Standard errors are clustered by firm. 60 Table 5: Internal Control and Exposed Corruption Cases Dependent variable=EXPOSED Predicted Signs Intercept ? ICFR − PAYRATIO − MGMTSHR − SALE + ROA − EISSUE + DISSUE + TOP1SHR − INVCOMP − BOARD − CEOCHAIR + MARKETIZATION − Year effects Industry effects No. of Obs. Pseudo R2 Full SOE Non-SOE -9.075*** (0.001) 0.701 (0.747) -0.132** (0.012) -0.142* (0.079) 0.293*** (0.003) -2.179 (0.150) -0.668 (0.975) 0.107 (0.206) 1.115 (0.854) -0.255 (0.364) 1.122 (0.665) 0.349 (0.184) -0.069 (0.179) Included Included -8.941*** (0.001) 0.509 (0.671) -0.101** (0.034) -0.115 (0.377) 0.329*** (0.001) -2.716 (0.131) -0.659 (0.967) 0.116 (0.207) 1.803 (0.936) -0.738 (0.182) 0.571 (0.587) 0.086 (0.420) -0.104 (0.103) Included Included 3.219 (0.718) -5.032* (0.083) -0.349*** (0.004) -0.648*** (0.004) -0.988 (0.984) 7.602 (0.993) 1.182 (0.116) 0.132 (0.417) -8.383** (0.045) 3.302 (0.862) 1.803 (0.658) 1.262** (0.019) 0.411 (0.940) Included Included 2,389 0.105 1,723 0.107 666 0.385 Test of coefficient difference SOEs vs. Non-SOEs Coefficient Diff. Z-statistic p-value ICFR 5.541* 1.589 0.056 All variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, and 0.10 level, respectively. P-values reported in parentheses are based on one-tailed tests for variables with predicted signs and two-tailed tests for variables without predicted signs. Standard errors are clustered by firm. 61 Table 6: Corporate Corruption and Future Operating Performance Dependent variable= Future Operating Performance (OIt+1) Predicted Column Column Signs A B 0.004 -0.001 Intercept (0.848) (0.945) – -0.106*** OTHREC (0.001) – -0.085** RELATEDLOAN (0.011) – PRIVATE EXPOSED – PPE ? GROWTH ? SEGMENTS ? EXPORT ? AGE ? BIG4 ? OI ? ASSETS ? TOP1SHR ? Year effects Industry effects No. of Obs. Adjusted R2 Column C -0.022 (0.517) Column D 0.014 (0.517) -0.001 (0.483) -0.002** (0.019) -0.002* (0.056) 0.002 (0.428) -0.003 (0.144) -0.003 (0.320) 0.008** (0.043) 0.583*** (0.000) 0.002* (0.076) 0.016** (0.022) Included Included -0.002** (0.044) -0.002* (0.050) 0.001 (0.471) -0.003 (0.158) -0.003 (0.297) 0.009** (0.026) 0.589*** (0.000) 0.002 (0.112) 0.018*** (0.009) Included Included -0.003* (0.092) -0.004* (0.099) 0.005* (0.088) -0.008** (0.020) -0.011*** (0.009) 0.002 (0.689) 0.608*** (0.000) 0.004 (0.111) 0.015 (0.171) Included Included -0.008** (0.029) -0.004*** (0.002) -0.003 (0.158) 0.0002 (0.920) -0.003 (0.182) 0.002 (0.434) 0.008* (0.056) 0.597*** (0.000) 0.002 (0.175) 0.004 (0.606) Included Included 4,767 0.406 4,767 0.406 2,124 0.410 2,795 0.433 PPE is the natural logarithm of gross property, plant and equipment at the end of year t. GROWTH is annual percentage sales growth in year t. AGE is the natural logarithm of the number of years that a company is publicly traded. All other variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. P-values reported in parentheses are based on one-tailed tests for variables with predicted signs and two-tailed tests for variables without predicted signs. Standard errors are clustered by firm. 62 Table 7: Ownership Structure and the Effectiveness of Internal Control Panel A: Full Sample Predicted sign OTHREC RELATEDLOAN PRIVATE EXPOSED ICFR − 0.003 -0.005 -0.008 -0.273 (0.608) (0.322) (0.341) (0.413) ICFR×TOP2to5IND − -0.030** -0.030** -0.034 1.830 (0.016) (0.025) (0.145) (0.851) TOP2to5IND − 0.013 0.015 0.016 -0.584 (0.972) (0.975) (0.842) (0.255) Control variables Included Included Included Included # Obs. 4722 4,722 1,796 2.389 0.140 0.054 0.049 0.109 Adjusted (Pseudo)R2 Panel B: SOEs versus Non-SOEs OTHREC RELATEDLOAN PRIVATE SOEs Non-SOEs SOEs Non-SOEs SOEs Non-SOEs ICFR -0.005 0.013 -0.000 -0.010 -0.007 -0.015 (0.309) (0.753) (0.491) (0.337) (0.376) (0.375) ICFR×TOP2to5IND -0.003 -0.076*** -0.012 -0.065** 0.011 -0.088* (0.427) (0.003) (0.202) (0.023) (0.342) (0.098) TOP2to5IND 0.001 0.032 0.008 0.027 -0.006 0.045 (0.547) (0.993) (0.868) (0.945) (0.330) (0.899) Control variables Included Included Included Included Included Included # Obs. 3,154 1,568 3,154 1,568 1,235 561 0.131 0.153 0.049 0.073 0.040 0.062 Adjusted (Pseudo)R2 EXPOSED SOEs Non-SOEs -0.483 -0.590 (0.358) (0.445) 1.952 -6.154* (0.838) (0.096) -0.593 2.201 (0.275) (0.832) Included Included 1,723 666 0.112 0.394 Test of coefficient difference ICFR×TOP2to5IND OTHREC SOEs vs. Non-SOEs Coefficient Diff. p-value 0.073*** 0.008 RELATEDLOAN SOEs vs. Non-SOEs Coefficient Diff. p-value 0.053* 0.066 63 PRIVATE SOEs vs. Non-SOEs Coefficient Diff. p-value 0.099* 0.088 EXPOSED SOEs vs. Non-SOEs Coefficient Diff. p-value 8.106* 0.056 Table 8: Internal Control and CEO Promotions Dependent variable=PROMOTION Predicted Full Signs ICFR + SALES ? OCF ? ROA + LEV ? MB ? CEOAGE ? EDUCATION + TENURE + TOP1SHR ? MARKETIZATION + Constant1 ? Constant 2 ? Constant ? Year effects Industry effects No. of Obs. Pseudo R2 Alla Turnover Onlyb NonSOEs Sample SOEs 0.915** 1.019** 0.752 (0.016) (0.026) (0.155) 0.090*** 0.082** 0.139*** (0.003) (0.033) (0.009) 0.230 -0.494 1.566** (0.608) (0.412) (0.022) 0.762 -0.206 1.214* (0.115) (0.885) (0.092) -0.306** -0.713*** -0.127 (0.015) (0.000) (0.464) -0.004 0.017 -0.025** (0.597) (0.113) (0.039) -0.261** -0.476*** 0.265 (0.039) (0.002) (0.263) 0.025 0.021 0.034 (0.309) (0.375) (0.327) 0.011 0.027 -0.022 (0.274) (0.122) (0.756) 0.053 0.258 -0.398 (0.846) (0.421) (0.475) 0.007 0.023 -0.029 (0.358) (0.182) (0.790) 0.440 0.345 0.995 (0.497) (0.686) (0.391) 5.366*** 5.324*** 5.979*** (0.000) (0.000) (0.000) Included Included Included Included Included Included 4,573 0.012 3,071 0.017 1,502 0.027 Full Sample 1.656** (0.025) 0.136** (0.030) 0.332 (0.705) -0.309 (0.608) -0.432 (0.174) -0.001 (0.939) -0.461* (0.076) 0.082 (0.208) 0.033 (0.204) 0.081 (0.886) 0.015 (0.357) SOEs 1.552* (0.071) 0.133 (0.108) -1.635 (0.170) -0.087 (0.523) -1.022** (0.033) 0.032* (0.089) -0.783** (0.011) 0.088 (0.262) 0.076* (0.066) 0.435 (0.529) 0.035 (0.249) NonSOEs 1.448 (0.177) 0.232** (0.045) 2.918* (0.064) 0.151 (0.465) -0.301 (0.450) -0.034 (0.173) 0.352 (0.471) -0.051 (0.618) -0.091 (0.866) -0.645 (0.572) -0.097 (0.920) -4.549*** -4.793*** -4.635* (0.000) (0.003) (0.056) Included Included Included Included Included Included 772 0.040 506 0.063 266 0.083 a : Includes all firm-years; the dependent variable equals 1 if the CEO is promoted in the following year, 0 if there is no CEO turnover, and –1 if the CEO is demoted in the following year. b : Includes only firm-years of CEO turnover; the dependent variable equals 1 if the CEO gets promoted in the following year, and 0 otherwise. We collect the CEO turnover data from 2008-2011 and eliminate cases where the CEO left office due to retirement or health reason, and cases where we cannot identify her next appointment. We consider the CEO as have been promoted if in the following year she left the CEO position and became (1) the chairman of the board at the same company (2) the CEO or the chairman of the board in the parent company, (3) the CEO at a larger public company, where the assets of the new company are at least 20% larger than the former company, or (4) a high-ranking government official. OCF is the operating cash flows divided by end-of-year total assets. LEV is leverage calculated by total liabilities divided by total assets. EDUCATION is a categorical variable measuring the highest degree of education the CEO holds; EDUCATION equals 4 if the CEO’s highest degree is a doctorate, 3 if it is a master’s degree, 2 if it is a college degrees, 1 if it is a 3-year college or other. TENURE measures the number of years that the CEO has assumed the position. CEOAGE is an indicator variable that equals to one if the CEO is 55 years old or older, and zero otherwise. All other variables are defined in Appendix C. All continuous variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively. P-values reported in parentheses are based on two-tailed tests. Standard errors are clustered by firm. 64 Table 9: Internal Control and Corporate Corruption: 2SLS Panel A: Full Sample Predicted sign OTHREC − -0.103** ܴܨܥܫ (0.029) Included Control variables 4722 # Obs. 0.140 Adjusted (Pseudo)R2 RELATEDLOAN -0.139*** (0.007) Included 4,722 0.053 PRIVATE -0.272*** (0.000) Included 1,796 0.134 EXPOSED 1.423 (0.615) Included 2.389 0.106 Panel B: SOEs versus Non-SOEs OTHREC SOEs 0.050 (0.847) Included 3,154 0.131 ܴܨܥܫ Control variables # Obs. Adjusted (Pseudo)R2 RELATEDLOAN Non-SOEs -0.226** (0.031) Included 1,568 0.148 SOEs -0.031 (0.229) Included 3,154 0.048 Non-SOEs -0.348*** (0.008) Included 1,568 0.069 PRIVATE SOEs -0.192*** (0.000) Included 1,235 0.087 EXPOSED Non-SOEs -0.426*** (0.002) Included 561 0.187 SOEs 1.020 (0.577) Included 1,723 0.110 Non-SOEs -46.335** (0.013) Included 666 0.438 Test of coefficient difference OTHREC SOEs vs. Non-SOEs ܴܨܥܫ Coefficient Diff. 0.275** p-value 0.017 RELATEDLOAN SOEs vs. Non-SOEs Coefficient Diff. 0.318** p-value 0.017 PRIVATE SOEs vs. Non-SOEs Coefficient Diff. 0.234* p-value 0.061 EXPOSED SOEs vs. Non-SOEs Coefficient Diff. 47.354** p-value 0.014 , is the predicted value of ICFR from the first-stage regression (see Appendix D). All other variables are defined in Appendix C. All continuous The variable of interest, ܴܨܥܫ variables are winsorized at 1% and 99% to mitigate outliers. ***, **, * indicate statistical significance at the 0.01, 0.05, 0.10 level, respectively, under one-tailed tests for variables with predicted signs and two-tailed tests for variables without predicted signs. Standard errors are clustered by firm. 65
© Copyright 2026 Paperzz